Best Way to Learn About AI | VibeBuilders.ai Resource Directory

[N] How Stability AI’s Founder Tanked His Billion-Dollar Startup

[N] How Stability AI’s Founder Tanked His Billion-Dollar Startup

forbes article: https://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup/ archive no paywall: https://archive.is/snbeV How Stability AI’s Founder Tanked His Billion-Dollar Startup Mar 29, 2024 Stability AI founder Emad Mostaque took the stage last week at the Terranea Resort in Palos Verdes, California to roaring applause and an introduction from an AI-generated Aristotle who announced him as “a modern Prometheus” with “the astuteness of Athena and the vision of Daedalus.” “Under his stewardship, AI becomes the Herculean force poised to vanquish the twin serpents of illness and ailment and extend the olive branch of longevity,” the faux Aristotle proclaimed. “I think that’s the best intro I’ve ever had,” Mostaque said. But behind Mostaque's hagiographic introduction lay a grim and fast metastasizing truth. Stability, once one of AI’s buzziest startups, was floundering. It had been running out of money for months and Mostaque had been unable to secure enough additional funding. It had defaulted on payments to Amazon whose cloud service undergirded Stability’s core offerings. The star research team behind its flagship text-to-image generator Stable Diffusion had tendered their resignations just three days before — as Forbes would first report — and other senior leaders had issued him an ultimatum: resign, or we walk too. Still, onstage before a massive audience of peers and acolytes, Mostaque talked a big game. “AI is jet planes for the mind,” he opined. “AI is our collective intelligence. It's the human Colossus.” He claimed a new, faster version of the Stable Diffusion image generator released earlier this month could generate “200 cats with hats per second.” But later, when he was asked about Stability’s financial model, Mostaque fumbled. “I can’t say that publicly,” he replied. “But it’s going well. We’re ahead of forecast.” Four days later, Mostaque stepped down as CEO of Stability, as Forbes first reported. In a post to X, the service formerly known as Twitter, he claimed he’d voluntarily abdicated his role to decentralize “the concentration of power in AI.” But sources told Forbes that was hardly the case. Behind the scenes, Mostaque had fought to maintain his position and control despite mounting pressure externally and internally to step down. Company documents and interviews with 32 current and former employees, investors, collaborators and industry observers suggest his abrupt exit was the result of poor business judgment and wild overspending that undermined confidence in his vision and leadership, and ultimately kneecapped the company. Mostaque, through his attorneys, declined to comment on record on a detailed list of questions about the reporting in this story. But in an email to Forbes earlier this week he broadly disputed the allegations. “Nobody tells you how hard it is to be a CEO and there are better CEOs than me to scale a business,” he said in a statement. “I am not sure anyone else would have been able to build and grow the research team to build the best and most widely used models out there and I’m very proud of the team there. I look forward to moving onto the next problem to handle and hopefully move the needle.” In an emailed statement, Christian Laforte and Shan Shan Wong, the interim co-CEOs who replaced Mostaque, said, "the company remains focused on commercializing its world leading technology” and providing it “to partners across the creative industries." After starting Stability in 2019, Mostaque built the company into an early AI juggernaut by seizing upon a promising research project that would become Stable Diffusion and funding it into a business reality. The ease with which the software generated detailed images from the simplest text prompts immediately captivated the public: 10 million people used it on any given day, the company told Forbes in early 2023. For some true believers, Mostaque was a crucial advocate for open-source AI development in a space dominated by the closed systems of OpenAI, Google and Anthropic. But his startup’s rise to one of the buzziest in generative AI was in part built on a series of exaggerations and misleading claims, as Forbes first reported last year (Mostaque disputed some points at the time). And they continued after he raised $100 million at a $1 billion valuation just days after launching Stable Diffusion in 2022. His failure to deliver on an array of grand promises, like building bespoke AI models for nation states, and his decision to pour tens of millions into research without a sustainable business plan, eroded Stability’s foundations and jeopardized its future. "He was just giving shit away,” one former employee told Forbes. “That man legitimately wanted to transform the world. He actually wanted to train AI models for kids in Malawi. Was it practical? Absolutely not." By October 2023, Stability would have less than $4 million left in the bank, according to an internal memo prepared for a board meeting and reviewed by Forbes. And mounting debt, including months of overdue Amazon Web Services payments, had already left it in the red. To avoid legal penalties for skipping Americans staff’s payroll, the document explained, the London-based startup was considering delaying tax payments to the U.K. government. It was Stability’s armada of GPUs, the wildly powerful and equally expensive chips undergirding AI, that were so taxing the company’s finances. Hosted by AWS, they had long been one of Mostaque’s bragging points; he often touted them as one of the world’s 10 largest supercomputers. They were responsible for helping Stability’s researchers build and maintain one of the top AI image generators, as well as break important new ground on generative audio, video and 3D models. “Undeniably, Stability has continued to ship a lot of models,” said one former employee. “They may not have profited off of it, but the broader ecosystem benefitted in a huge, huge way.” But the costs associated with so much compute were now threatening to sink the company. According to an internal October financial forecast seen by Forbes, Stability was on track to spend $99 million on compute in 2023. It noted as well that Stability was “underpaying AWS bills for July (by $1M)” and “not planning to pay AWS at the end of October for August usage ($7M).” Then there were the September and October bills, plus $1 million owed to Google Cloud and $600,000 to GPU cloud data center CoreWeave. (Amazon, Google and CoreWeave declined to comment.) With an additional $54 million allocated to wages and operating expenses, Stability’s total projected costs for 2023 were $153 million. But according to its October financial report, its projected revenue for the calendar year was just $11 million. Stability was on track to lose more money per month than it made in an entire year. The company’s dire financial position had thoroughly soured Stability’s current investors, including Coatue, which had invested tens of millions in the company during its $101 million funding round in 2022. In the middle of 2023, Mostaque agreed to an independent audit after Coatue raised a series of concerns, according to a source with direct knowledge of the matter. The outcome of the investigation is unclear. Coatue declined to comment. Within a week of an early October board meeting where Mostaque shared that financial forecast, Lightspeed Venture Partners, another major investor, sent a letter to the board urging them to sell the company. The distressing numbers had “severely undermined” the firm’s confidence in Mostaque’s ability to lead the company. “In particular, we are surprised and deeply concerned by a cash position just now disclosed to us that is inconsistent with prior discussions on this topic,” Lightspeed’s general counsel Brett Nissenberg wrote in the letter, a copy of which was viewed by Forbes. “Lightspeed believes that the company is not likely financeable on terms that would assure the company’s long term sound financial position.” (Lightspeed declined a request for comment.) The calls for a sale led Stability to quietly begin looking for a buyer. Bloomberg reported in November that Stability approached AI startups Cohere and Jasper to gauge their interest. Stability denied this, and Jasper CEO Timothy Young did the same when reached for comment by Forbes. A Cohere representative declined to comment. But one prominent AI company confirmed that Mostaque’s representatives had reached out to them to test the waters. Those talks did not advance because “the numbers didn’t add up,” this person, who declined to be named due to the confidential nature of the talks, told Forbes. Stability also tried to court Samsung as a buyer, going so far as to redecorate its office in advance of a planned meeting with the Korean electronics giant. (Samsung said that it invested in Stability in 2023 and that it does not comment on M&A discussions.) Coatue had been calling for Mostaque’s resignation for months, according to a source with direct knowledge. But it and other investors were unable to oust him because he was the company’s majority shareholder. When they tried a different tact by rallying other investors to offer him a juicy equity package to resign, Mostaque refused, said two sources. By October, Coatue and Lightspeed had had enough. Coatue left the board and Lightspeed resigned its observer seat. “Emad infuriated our initial investors so much it’s just making it impossible for us to raise more money under acceptable terms,” one current Stability executive told Forbes. The early months of 2024 saw Stability’s already precarious position eroding further still. Employees were quietly laid off. Three people in a position to know estimated that at least 10% of staff were cut. And cash reserves continued to dwindle. Mostaque mentioned a lifeline at the October board meeting: $95 million in tentative funding from new investors, pending due diligence. But in the end, only a fraction of it was wired, two sources say, much of it from Intel, which Forbes has learned invested $20 million, a fraction of what was reported. (Intel did not return a request for comment by publication time.) Two hours after Forbes broke the news of Mostaque’s plans to step down as CEO, Stability issued a press release confirming his resignation. Chief operating officer Wong and chief technology officer Laforte have taken over in the interim. Mostaque, who said on X that he still owns a majority of the company, also stepped down from the board, which has now initiated a search for a permanent CEO. There is a lot of work to be done to turn things around, and very little time in which to do it. Said the current Stability executive, “There’s still a possibility of a turnaround story, but the odds drop by the day.” In July of 2023, Mostaque still thought he could pull it off. Halfway through the month, he shared a fundraising plan with his lieutenants. It was wildly optimistic, detailing the raise of $500 million in cash and another $750 million in computing facilities from marquee investors like Nvidia, Google, Intel and the World Bank (Nvidia and Google declined comment. Intel did not respond. The World Bank said it did not invest in Stability). In a Slack message reviewed by Forbes, Mostaque said Google was “willing to move fast” and the round was “likely to be oversubscribed.” It wasn’t. Three people with direct knowledge of these fundraising efforts told Forbes that while there was some interest in Stability, talks often stalled when it came time to disclose financials. Two of them noted that earlier in the year, Mostaque had simply stopped engaging with VCs who asked for numbers. Only one firm invested around that time: actor Ashton Kutcher’s Sound Ventures, which invested $35 million in the form of a convertible SAFE note during the second quarter, according to an internal document. (Sound Ventures did not respond to a request for comment.) And though he’d managed to score a meeting with Nvidia and its CEO Jensen Huang, it ended in disaster, according to two sources. “Under Jensen's microscopic questions, Emad just fell apart,” a source in position to know told Forbes. Huang quickly concluded Stability wasn’t ready for an investment from Nvidia, the sources said. Mostaque told Forbes in an email that he had not met with Huang since 2022, except to say “hello and what’s up a few times after.” His July 2023 message references a plan to raise $150 million from Nvidia. (Nvidia declined to comment.) After a June Forbes investigation citing more than 30 sources revealed Mostaque’s history of misleading claims, Mostaque struggled to raise funding, a Stability investor told Forbes. (Mostaque disputed the story at the time and called it "coordinated lies" in his email this week to Forbes). Increasingly, investors scrutinized his assertions and pressed for data. And Young, now the CEO of Jasper, turned down a verbal offer to be Stability’s president after reading the article, according to a source with direct knowledge of the matter. The collapse of the talks aggravated the board and other executives, who had hoped Young would compensate for the sales and business management skills that Mostaque lacked, according to four people in a position to know. (Young declined to comment.) When Stability’s senior leadership convened in London for the CogX conference in September, the financing had still not closed. There, a group of executives confronted Mostaque asking questions about the company’s cash position and runway, according to three people with direct knowledge of the incident. They did not get the clarity they’d hoped for. By October, Mostaque had reduced his fundraising target by more than 80%. The months that followed saw a steady drumbeat of departures — general counsel Adam Avrunin, vice presidents Mike Melnicki, Ed Newton-Rex and Joe Penna, chief people officer Ozden Onder — culminating in the demoralizing March exit of Stable Diffusion’s primary developers Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz. Rombach, who led the team, had been angling to leave for months, two sources said, first threatening to resign last summer because of the fundraising failures. Others left over concerns about cash flow, as well as liabilities — including what four people described as Mostaque’s lax approach to ensuring that Stability products could not be used to produce child sexual abuse imagery. “Stability AI is committed to preventing the misuse of AI and prohibits the use of our image models and services for unlawful activity, including attempts to edit or create CSAM,” Ella Irwin, senior vice president of integrity, said in a statement. Newton-Rex told Forbes he resigned because he disagreed with Stability’s position that training AI on copyrighted work without consent is fair use. Melnicki and Penna declined to comment. Avrunin and Onder could not be reached for comment. None of the researchers responded to requests for comment. The Stable Diffusion researchers’ departure as a cohort says a lot about the state of Stability AI. The company’s researchers were widely viewed as its crown jewels, their work subsidized with a firehose of pricey compute power that was even extended to people outside the company. Martino Russi, an artificial intelligence researcher, told Forbes that though he was never formally employed by Stability, the company provided him a “staggering” amount of compute between January and April 2023 to play around with developing an AI video generator that Stability might someday use. “It was Candy Land or Coney Island,” said Russi, who estimates that his experiment, which was ultimately shelved, cost the company $2.5 million. Stable Diffusion was simultaneously Stability’s marquee product and its existential cash crisis. One current employee described it to Forbes as “a giant vacuum that absorbed everything: money, compute, people.” While the software was widely used, with Mostaque claiming downloads reaching into the hundreds of millions, Stability struggled to translate that wild success into revenue. Mostaque knew it could be done — peers at Databricks, Elastic and MongoDB had all turned a free product into a lucrative business — he just couldn’t figure out how. His first attempt was Stability’s API, which allowed paying customers to integrate Stable Diffusion into their own products. In early 2023, a handful of small companies, like art generator app NightCafe and presentation software startup Tome, signed on, according to four people with knowledge of the deals. But Stability’s poor account management services soured many, and in a matter of months NightCafe and Tome canceled their contracts, three people said. NightCafe founder Angus Russell told Forbes that his company switched to a competitor which “offered much cheaper inference costs and a broader service.” Tome did not respond to a request for comment. Meanwhile, Mostaque’s efforts to court larger companies like Samsung and Snapchat were failing, according to five people familiar with the effort. Canva, which was already one of the heaviest users of open-sourced Stable Diffusion, had multiple discussions with Stability, which was angling for a contract it hoped would generate several millions in annual revenue. But the deal never materialized, four sources said. “These three companies wanted and needed us,” one former employee told Forbes. “They would have been the perfect customers.” (Samsung, Snap and Canva declined to comment.) “It’s not that there was not an appetite to pay Stability — there were tons of companies that would have that wanted to,” the former employee said. “There was a huge opportunity and demand, but just a resistance to execution.” Mostaque’s other big idea was to provide governments with bespoke national AI models that would invigorate their economies and citizenry. “Emad envisions a world where AI through 100 national models serves not as a tool of the few, but as a benefactor to all promising to confront great adversaries, cancer, autism, and the sands of time itself,” the AI avatar of Aristotle said in his intro at the conference. Mostaque told several prospective customers that he could deliver such models within 60 days — an untenable timeline, according to two people in position to know. Stability attempted to develop a model for the Singaporean government over the protestation of employees who questioned its technical feasibility, three sources familiar with the effort told Forbes. But it couldn’t pull it off and Singapore never became a customer. (The government of Singapore confirmed it did not enter into a deal with Stability, but declined to answer additional questions.) As Stability careened from one new business idea to another, resources were abruptly reallocated and researchers reassigned. The whiplash shifts in a largely siloed organization demoralized and infuriated employees. “There were ‘urgent’ things, ‘urgent urgent’ things and ‘most urgent,’” one former employee complained. “None of these things seem important if everything is important.” Another former Stability executive was far more pointed in their assessment. “Emad is the most disorganized leader I have ever worked with in my career,” this person told Forbes. “He has no vision, and changes directions every week, often based on what he sees on Twitter.” In a video interview posted shortly before this story was published, Mostaque explained his leadership style: “I'm particularly great at taking creatives, developers, researchers, others, and achieving their full potential in designing systems. But I should not be dealing with, you know, HR and operations and business development and other elements. There are far better people than me to do that.” By December 2023, Stability had partially abandoned its open-source roots and announced that any commercial use of Stable Diffusion would cost customers at least $20 per month (non-commercial and research use of Stable Diffusion would remain free). But privately, Stability was considering a potentially more lucrative source of revenue: reselling the compute it was leasing from providers like AWS, according to six people familiar with the effort. Though it was essentially GPU arbitrage, Stability framed the strategy to investors as a “managed services” offering. Its damning October financial report projected optimistically that such an offering would bring in $139 million in 2024 — 98% of its revenue. Multiple employees at the time told Forbes they feared reselling compute, even if the company called it “managed services,” would violate the terms of Stability’s contract with AWS. Amazon declined to comment. “The line internally was that we are not reselling compute,” one former employee said. “This was some of the dirtiest feeling stuff.” Stability also discussed reselling a cluster of Nvidia A100 chips, leased via CoreWeave, to the venture capital firm Andreessen Horowitz, three sources said. “It was under the guise of managed services, but there wasn’t any management happening,” one of these people told Forbes. Andreessen Horowitz and CoreWeave declined to comment. Stability did not respond to questions about if it plans to continue this strategy now that Mostaque is out of the picture. Regardless, interim co-CEOs Wong and Laforte are on a tight timeline to clean up his mess. Board chairman Jim O’Shaughnessy said in a statement that he was confident the pair “will adeptly steer the company forward in developing and commercializing industry-leading generative AI products.” But burn continues to far outpace revenue. The Financial Times reported Friday that the company made $5.4 million of revenue in February, against $8 million in costs. Several sources said there are ongoing concerns about making payroll for the roughly 150 remaining employees. Leadership roles have gone vacant for months amid the disarray, leaving the company increasingly directionless. Meanwhile, a potentially catastrophic legal threat looms over the company: A trio of copyright infringement lawsuits brought by Getty Images and a group of artists in the U.S. and U.K., who claim Stability illegally used their art and photography to train the AI models powering Stable Diffusion. A London-based court has already rejected the company’s bid to throw out one of the lawsuits on the basis that none of its researchers were based in the U.K. And Stability’s claim that Getty’s Delaware lawsuit should be blocked because it's a U.K.-based company was rejected. (Stability did not respond to questions about the litigation.) AI-related copyright litigation “could go on for years,” according to Eric Goldman, a law professor at Santa Clara University. He told Forbes that though plaintiffs suing AI firms face an uphill battle overcoming the existing legal precedent on copyright infringement, the quantity of arguments available to make are virtually inexhaustible. “Like in military theory, if there’s a gap in your lines, that’s where the enemy pours through — if any one of those arguments succeeds, it could completely change the generative AI environment,” he said. “In some sense, generative AI as an industry has to win everything.” Stability, which had more than $100 million in the bank just a year and a half ago, is in a deep hole. Not only does it need more funding, it needs a viable business model — or a buyer with the vision and chops to make it successful in a fast-moving and highly competitive sector. At an all hands meeting this past Monday, Stability’s new leaders detailed a path forward. One point of emphasis: a plan to better manage resources and expenses, according to one person in attendance. It’s a start, but Mostaque’s meddling has left them with little runway to execute. His resignation, though, has given some employees hope. “A few people are 100% going to reconsider leaving after today,” said one current employee. “And the weird gloomy aura of hearing Emad talking nonsense for an hour is gone.” Shortly before Mostaque resigned, one current Stability executive told Forbes that they were optimistic his departure could make Stability appealing enough to receive a small investment or sale to a friendly party. “There are companies that have raised hundreds of millions of dollars that have much less intrinsic value than Stability,” the person said. “A white knight may still appear.”

#reddit #stability #ai’s

0

0.667

milaworld•This week

[N] How Stability AI’s Founder Tanked His Billion-Dollar Startup

forbes article: https://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup/ archive no paywall: https://archive.is/snbeV How Stability AI’s Founder Tanked His Billion-Dollar Startup Mar 29, 2024 Stability AI founder Emad Mostaque took the stage last week at the Terranea Resort in Palos Verdes, California to roaring applause and an introduction from an AI-generated Aristotle who announced him as “a modern Prometheus” with “the astuteness of Athena and the vision of Daedalus.” “Under his stewardship, AI becomes the Herculean force poised to vanquish the twin serpents of illness and ailment and extend the olive branch of longevity,” the faux Aristotle proclaimed. “I think that’s the best intro I’ve ever had,” Mostaque said. But behind Mostaque's hagiographic introduction lay a grim and fast metastasizing truth. Stability, once one of AI’s buzziest startups, was floundering. It had been running out of money for months and Mostaque had been unable to secure enough additional funding. It had defaulted on payments to Amazon whose cloud service undergirded Stability’s core offerings. The star research team behind its flagship text-to-image generator Stable Diffusion had tendered their resignations just three days before — as Forbes would first report — and other senior leaders had issued him an ultimatum: resign, or we walk too. Still, onstage before a massive audience of peers and acolytes, Mostaque talked a big game. “AI is jet planes for the mind,” he opined. “AI is our collective intelligence. It's the human Colossus.” He claimed a new, faster version of the Stable Diffusion image generator released earlier this month could generate “200 cats with hats per second.” But later, when he was asked about Stability’s financial model, Mostaque fumbled. “I can’t say that publicly,” he replied. “But it’s going well. We’re ahead of forecast.” Four days later, Mostaque stepped down as CEO of Stability, as Forbes first reported. In a post to X, the service formerly known as Twitter, he claimed he’d voluntarily abdicated his role to decentralize “the concentration of power in AI.” But sources told Forbes that was hardly the case. Behind the scenes, Mostaque had fought to maintain his position and control despite mounting pressure externally and internally to step down. Company documents and interviews with 32 current and former employees, investors, collaborators and industry observers suggest his abrupt exit was the result of poor business judgment and wild overspending that undermined confidence in his vision and leadership, and ultimately kneecapped the company. Mostaque, through his attorneys, declined to comment on record on a detailed list of questions about the reporting in this story. But in an email to Forbes earlier this week he broadly disputed the allegations. “Nobody tells you how hard it is to be a CEO and there are better CEOs than me to scale a business,” he said in a statement. “I am not sure anyone else would have been able to build and grow the research team to build the best and most widely used models out there and I’m very proud of the team there. I look forward to moving onto the next problem to handle and hopefully move the needle.” In an emailed statement, Christian Laforte and Shan Shan Wong, the interim co-CEOs who replaced Mostaque, said, "the company remains focused on commercializing its world leading technology” and providing it “to partners across the creative industries." After starting Stability in 2019, Mostaque built the company into an early AI juggernaut by seizing upon a promising research project that would become Stable Diffusion and funding it into a business reality. The ease with which the software generated detailed images from the simplest text prompts immediately captivated the public: 10 million people used it on any given day, the company told Forbes in early 2023. For some true believers, Mostaque was a crucial advocate for open-source AI development in a space dominated by the closed systems of OpenAI, Google and Anthropic. But his startup’s rise to one of the buzziest in generative AI was in part built on a series of exaggerations and misleading claims, as Forbes first reported last year (Mostaque disputed some points at the time). And they continued after he raised $100 million at a $1 billion valuation just days after launching Stable Diffusion in 2022. His failure to deliver on an array of grand promises, like building bespoke AI models for nation states, and his decision to pour tens of millions into research without a sustainable business plan, eroded Stability’s foundations and jeopardized its future. "He was just giving shit away,” one former employee told Forbes. “That man legitimately wanted to transform the world. He actually wanted to train AI models for kids in Malawi. Was it practical? Absolutely not." By October 2023, Stability would have less than $4 million left in the bank, according to an internal memo prepared for a board meeting and reviewed by Forbes. And mounting debt, including months of overdue Amazon Web Services payments, had already left it in the red. To avoid legal penalties for skipping Americans staff’s payroll, the document explained, the London-based startup was considering delaying tax payments to the U.K. government. It was Stability’s armada of GPUs, the wildly powerful and equally expensive chips undergirding AI, that were so taxing the company’s finances. Hosted by AWS, they had long been one of Mostaque’s bragging points; he often touted them as one of the world’s 10 largest supercomputers. They were responsible for helping Stability’s researchers build and maintain one of the top AI image generators, as well as break important new ground on generative audio, video and 3D models. “Undeniably, Stability has continued to ship a lot of models,” said one former employee. “They may not have profited off of it, but the broader ecosystem benefitted in a huge, huge way.” But the costs associated with so much compute were now threatening to sink the company. According to an internal October financial forecast seen by Forbes, Stability was on track to spend $99 million on compute in 2023. It noted as well that Stability was “underpaying AWS bills for July (by $1M)” and “not planning to pay AWS at the end of October for August usage ($7M).” Then there were the September and October bills, plus $1 million owed to Google Cloud and $600,000 to GPU cloud data center CoreWeave. (Amazon, Google and CoreWeave declined to comment.) With an additional $54 million allocated to wages and operating expenses, Stability’s total projected costs for 2023 were $153 million. But according to its October financial report, its projected revenue for the calendar year was just $11 million. Stability was on track to lose more money per month than it made in an entire year. The company’s dire financial position had thoroughly soured Stability’s current investors, including Coatue, which had invested tens of millions in the company during its $101 million funding round in 2022. In the middle of 2023, Mostaque agreed to an independent audit after Coatue raised a series of concerns, according to a source with direct knowledge of the matter. The outcome of the investigation is unclear. Coatue declined to comment. Within a week of an early October board meeting where Mostaque shared that financial forecast, Lightspeed Venture Partners, another major investor, sent a letter to the board urging them to sell the company. The distressing numbers had “severely undermined” the firm’s confidence in Mostaque’s ability to lead the company. “In particular, we are surprised and deeply concerned by a cash position just now disclosed to us that is inconsistent with prior discussions on this topic,” Lightspeed’s general counsel Brett Nissenberg wrote in the letter, a copy of which was viewed by Forbes. “Lightspeed believes that the company is not likely financeable on terms that would assure the company’s long term sound financial position.” (Lightspeed declined a request for comment.) The calls for a sale led Stability to quietly begin looking for a buyer. Bloomberg reported in November that Stability approached AI startups Cohere and Jasper to gauge their interest. Stability denied this, and Jasper CEO Timothy Young did the same when reached for comment by Forbes. A Cohere representative declined to comment. But one prominent AI company confirmed that Mostaque’s representatives had reached out to them to test the waters. Those talks did not advance because “the numbers didn’t add up,” this person, who declined to be named due to the confidential nature of the talks, told Forbes. Stability also tried to court Samsung as a buyer, going so far as to redecorate its office in advance of a planned meeting with the Korean electronics giant. (Samsung said that it invested in Stability in 2023 and that it does not comment on M&A discussions.) Coatue had been calling for Mostaque’s resignation for months, according to a source with direct knowledge. But it and other investors were unable to oust him because he was the company’s majority shareholder. When they tried a different tact by rallying other investors to offer him a juicy equity package to resign, Mostaque refused, said two sources. By October, Coatue and Lightspeed had had enough. Coatue left the board and Lightspeed resigned its observer seat. “Emad infuriated our initial investors so much it’s just making it impossible for us to raise more money under acceptable terms,” one current Stability executive told Forbes. The early months of 2024 saw Stability’s already precarious position eroding further still. Employees were quietly laid off. Three people in a position to know estimated that at least 10% of staff were cut. And cash reserves continued to dwindle. Mostaque mentioned a lifeline at the October board meeting: $95 million in tentative funding from new investors, pending due diligence. But in the end, only a fraction of it was wired, two sources say, much of it from Intel, which Forbes has learned invested $20 million, a fraction of what was reported. (Intel did not return a request for comment by publication time.) Two hours after Forbes broke the news of Mostaque’s plans to step down as CEO, Stability issued a press release confirming his resignation. Chief operating officer Wong and chief technology officer Laforte have taken over in the interim. Mostaque, who said on X that he still owns a majority of the company, also stepped down from the board, which has now initiated a search for a permanent CEO. There is a lot of work to be done to turn things around, and very little time in which to do it. Said the current Stability executive, “There’s still a possibility of a turnaround story, but the odds drop by the day.” In July of 2023, Mostaque still thought he could pull it off. Halfway through the month, he shared a fundraising plan with his lieutenants. It was wildly optimistic, detailing the raise of $500 million in cash and another $750 million in computing facilities from marquee investors like Nvidia, Google, Intel and the World Bank (Nvidia and Google declined comment. Intel did not respond. The World Bank said it did not invest in Stability). In a Slack message reviewed by Forbes, Mostaque said Google was “willing to move fast” and the round was “likely to be oversubscribed.” It wasn’t. Three people with direct knowledge of these fundraising efforts told Forbes that while there was some interest in Stability, talks often stalled when it came time to disclose financials. Two of them noted that earlier in the year, Mostaque had simply stopped engaging with VCs who asked for numbers. Only one firm invested around that time: actor Ashton Kutcher’s Sound Ventures, which invested $35 million in the form of a convertible SAFE note during the second quarter, according to an internal document. (Sound Ventures did not respond to a request for comment.) And though he’d managed to score a meeting with Nvidia and its CEO Jensen Huang, it ended in disaster, according to two sources. “Under Jensen's microscopic questions, Emad just fell apart,” a source in position to know told Forbes. Huang quickly concluded Stability wasn’t ready for an investment from Nvidia, the sources said. Mostaque told Forbes in an email that he had not met with Huang since 2022, except to say “hello and what’s up a few times after.” His July 2023 message references a plan to raise $150 million from Nvidia. (Nvidia declined to comment.) After a June Forbes investigation citing more than 30 sources revealed Mostaque’s history of misleading claims, Mostaque struggled to raise funding, a Stability investor told Forbes. (Mostaque disputed the story at the time and called it "coordinated lies" in his email this week to Forbes). Increasingly, investors scrutinized his assertions and pressed for data. And Young, now the CEO of Jasper, turned down a verbal offer to be Stability’s president after reading the article, according to a source with direct knowledge of the matter. The collapse of the talks aggravated the board and other executives, who had hoped Young would compensate for the sales and business management skills that Mostaque lacked, according to four people in a position to know. (Young declined to comment.) When Stability’s senior leadership convened in London for the CogX conference in September, the financing had still not closed. There, a group of executives confronted Mostaque asking questions about the company’s cash position and runway, according to three people with direct knowledge of the incident. They did not get the clarity they’d hoped for. By October, Mostaque had reduced his fundraising target by more than 80%. The months that followed saw a steady drumbeat of departures — general counsel Adam Avrunin, vice presidents Mike Melnicki, Ed Newton-Rex and Joe Penna, chief people officer Ozden Onder — culminating in the demoralizing March exit of Stable Diffusion’s primary developers Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz. Rombach, who led the team, had been angling to leave for months, two sources said, first threatening to resign last summer because of the fundraising failures. Others left over concerns about cash flow, as well as liabilities — including what four people described as Mostaque’s lax approach to ensuring that Stability products could not be used to produce child sexual abuse imagery. “Stability AI is committed to preventing the misuse of AI and prohibits the use of our image models and services for unlawful activity, including attempts to edit or create CSAM,” Ella Irwin, senior vice president of integrity, said in a statement. Newton-Rex told Forbes he resigned because he disagreed with Stability’s position that training AI on copyrighted work without consent is fair use. Melnicki and Penna declined to comment. Avrunin and Onder could not be reached for comment. None of the researchers responded to requests for comment. The Stable Diffusion researchers’ departure as a cohort says a lot about the state of Stability AI. The company’s researchers were widely viewed as its crown jewels, their work subsidized with a firehose of pricey compute power that was even extended to people outside the company. Martino Russi, an artificial intelligence researcher, told Forbes that though he was never formally employed by Stability, the company provided him a “staggering” amount of compute between January and April 2023 to play around with developing an AI video generator that Stability might someday use. “It was Candy Land or Coney Island,” said Russi, who estimates that his experiment, which was ultimately shelved, cost the company $2.5 million. Stable Diffusion was simultaneously Stability’s marquee product and its existential cash crisis. One current employee described it to Forbes as “a giant vacuum that absorbed everything: money, compute, people.” While the software was widely used, with Mostaque claiming downloads reaching into the hundreds of millions, Stability struggled to translate that wild success into revenue. Mostaque knew it could be done — peers at Databricks, Elastic and MongoDB had all turned a free product into a lucrative business — he just couldn’t figure out how. His first attempt was Stability’s API, which allowed paying customers to integrate Stable Diffusion into their own products. In early 2023, a handful of small companies, like art generator app NightCafe and presentation software startup Tome, signed on, according to four people with knowledge of the deals. But Stability’s poor account management services soured many, and in a matter of months NightCafe and Tome canceled their contracts, three people said. NightCafe founder Angus Russell told Forbes that his company switched to a competitor which “offered much cheaper inference costs and a broader service.” Tome did not respond to a request for comment. Meanwhile, Mostaque’s efforts to court larger companies like Samsung and Snapchat were failing, according to five people familiar with the effort. Canva, which was already one of the heaviest users of open-sourced Stable Diffusion, had multiple discussions with Stability, which was angling for a contract it hoped would generate several millions in annual revenue. But the deal never materialized, four sources said. “These three companies wanted and needed us,” one former employee told Forbes. “They would have been the perfect customers.” (Samsung, Snap and Canva declined to comment.) “It’s not that there was not an appetite to pay Stability — there were tons of companies that would have that wanted to,” the former employee said. “There was a huge opportunity and demand, but just a resistance to execution.” Mostaque’s other big idea was to provide governments with bespoke national AI models that would invigorate their economies and citizenry. “Emad envisions a world where AI through 100 national models serves not as a tool of the few, but as a benefactor to all promising to confront great adversaries, cancer, autism, and the sands of time itself,” the AI avatar of Aristotle said in his intro at the conference. Mostaque told several prospective customers that he could deliver such models within 60 days — an untenable timeline, according to two people in position to know. Stability attempted to develop a model for the Singaporean government over the protestation of employees who questioned its technical feasibility, three sources familiar with the effort told Forbes. But it couldn’t pull it off and Singapore never became a customer. (The government of Singapore confirmed it did not enter into a deal with Stability, but declined to answer additional questions.) As Stability careened from one new business idea to another, resources were abruptly reallocated and researchers reassigned. The whiplash shifts in a largely siloed organization demoralized and infuriated employees. “There were ‘urgent’ things, ‘urgent urgent’ things and ‘most urgent,’” one former employee complained. “None of these things seem important if everything is important.” Another former Stability executive was far more pointed in their assessment. “Emad is the most disorganized leader I have ever worked with in my career,” this person told Forbes. “He has no vision, and changes directions every week, often based on what he sees on Twitter.” In a video interview posted shortly before this story was published, Mostaque explained his leadership style: “I'm particularly great at taking creatives, developers, researchers, others, and achieving their full potential in designing systems. But I should not be dealing with, you know, HR and operations and business development and other elements. There are far better people than me to do that.” By December 2023, Stability had partially abandoned its open-source roots and announced that any commercial use of Stable Diffusion would cost customers at least $20 per month (non-commercial and research use of Stable Diffusion would remain free). But privately, Stability was considering a potentially more lucrative source of revenue: reselling the compute it was leasing from providers like AWS, according to six people familiar with the effort. Though it was essentially GPU arbitrage, Stability framed the strategy to investors as a “managed services” offering. Its damning October financial report projected optimistically that such an offering would bring in $139 million in 2024 — 98% of its revenue. Multiple employees at the time told Forbes they feared reselling compute, even if the company called it “managed services,” would violate the terms of Stability’s contract with AWS. Amazon declined to comment. “The line internally was that we are not reselling compute,” one former employee said. “This was some of the dirtiest feeling stuff.” Stability also discussed reselling a cluster of Nvidia A100 chips, leased via CoreWeave, to the venture capital firm Andreessen Horowitz, three sources said. “It was under the guise of managed services, but there wasn’t any management happening,” one of these people told Forbes. Andreessen Horowitz and CoreWeave declined to comment. Stability did not respond to questions about if it plans to continue this strategy now that Mostaque is out of the picture. Regardless, interim co-CEOs Wong and Laforte are on a tight timeline to clean up his mess. Board chairman Jim O’Shaughnessy said in a statement that he was confident the pair “will adeptly steer the company forward in developing and commercializing industry-leading generative AI products.” But burn continues to far outpace revenue. The Financial Times reported Friday that the company made $5.4 million of revenue in February, against $8 million in costs. Several sources said there are ongoing concerns about making payroll for the roughly 150 remaining employees. Leadership roles have gone vacant for months amid the disarray, leaving the company increasingly directionless. Meanwhile, a potentially catastrophic legal threat looms over the company: A trio of copyright infringement lawsuits brought by Getty Images and a group of artists in the U.S. and U.K., who claim Stability illegally used their art and photography to train the AI models powering Stable Diffusion. A London-based court has already rejected the company’s bid to throw out one of the lawsuits on the basis that none of its researchers were based in the U.K. And Stability’s claim that Getty’s Delaware lawsuit should be blocked because it's a U.K.-based company was rejected. (Stability did not respond to questions about the litigation.) AI-related copyright litigation “could go on for years,” according to Eric Goldman, a law professor at Santa Clara University. He told Forbes that though plaintiffs suing AI firms face an uphill battle overcoming the existing legal precedent on copyright infringement, the quantity of arguments available to make are virtually inexhaustible. “Like in military theory, if there’s a gap in your lines, that’s where the enemy pours through — if any one of those arguments succeeds, it could completely change the generative AI environment,” he said. “In some sense, generative AI as an industry has to win everything.” Stability, which had more than $100 million in the bank just a year and a half ago, is in a deep hole. Not only does it need more funding, it needs a viable business model — or a buyer with the vision and chops to make it successful in a fast-moving and highly competitive sector. At an all hands meeting this past Monday, Stability’s new leaders detailed a path forward. One point of emphasis: a plan to better manage resources and expenses, according to one person in attendance. It’s a start, but Mostaque’s meddling has left them with little runway to execute. His resignation, though, has given some employees hope. “A few people are 100% going to reconsider leaving after today,” said one current employee. “And the weird gloomy aura of hearing Emad talking nonsense for an hour is gone.” Shortly before Mostaque resigned, one current Stability executive told Forbes that they were optimistic his departure could make Stability appealing enough to receive a small investment or sale to a friendly party. “There are companies that have raised hundreds of millions of dollars that have much less intrinsic value than Stability,” the person said. “A white knight may still appear.”

#reddit #stability #ai’s

0

0.667

milaworld•This week

[N] How Stability AI’s Founder Tanked His Billion-Dollar Startup

forbes article: https://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup/ archive no paywall: https://archive.is/snbeV How Stability AI’s Founder Tanked His Billion-Dollar Startup Mar 29, 2024 Stability AI founder Emad Mostaque took the stage last week at the Terranea Resort in Palos Verdes, California to roaring applause and an introduction from an AI-generated Aristotle who announced him as “a modern Prometheus” with “the astuteness of Athena and the vision of Daedalus.” “Under his stewardship, AI becomes the Herculean force poised to vanquish the twin serpents of illness and ailment and extend the olive branch of longevity,” the faux Aristotle proclaimed. “I think that’s the best intro I’ve ever had,” Mostaque said. But behind Mostaque's hagiographic introduction lay a grim and fast metastasizing truth. Stability, once one of AI’s buzziest startups, was floundering. It had been running out of money for months and Mostaque had been unable to secure enough additional funding. It had defaulted on payments to Amazon whose cloud service undergirded Stability’s core offerings. The star research team behind its flagship text-to-image generator Stable Diffusion had tendered their resignations just three days before — as Forbes would first report — and other senior leaders had issued him an ultimatum: resign, or we walk too. Still, onstage before a massive audience of peers and acolytes, Mostaque talked a big game. “AI is jet planes for the mind,” he opined. “AI is our collective intelligence. It's the human Colossus.” He claimed a new, faster version of the Stable Diffusion image generator released earlier this month could generate “200 cats with hats per second.” But later, when he was asked about Stability’s financial model, Mostaque fumbled. “I can’t say that publicly,” he replied. “But it’s going well. We’re ahead of forecast.” Four days later, Mostaque stepped down as CEO of Stability, as Forbes first reported. In a post to X, the service formerly known as Twitter, he claimed he’d voluntarily abdicated his role to decentralize “the concentration of power in AI.” But sources told Forbes that was hardly the case. Behind the scenes, Mostaque had fought to maintain his position and control despite mounting pressure externally and internally to step down. Company documents and interviews with 32 current and former employees, investors, collaborators and industry observers suggest his abrupt exit was the result of poor business judgment and wild overspending that undermined confidence in his vision and leadership, and ultimately kneecapped the company. Mostaque, through his attorneys, declined to comment on record on a detailed list of questions about the reporting in this story. But in an email to Forbes earlier this week he broadly disputed the allegations. “Nobody tells you how hard it is to be a CEO and there are better CEOs than me to scale a business,” he said in a statement. “I am not sure anyone else would have been able to build and grow the research team to build the best and most widely used models out there and I’m very proud of the team there. I look forward to moving onto the next problem to handle and hopefully move the needle.” In an emailed statement, Christian Laforte and Shan Shan Wong, the interim co-CEOs who replaced Mostaque, said, "the company remains focused on commercializing its world leading technology” and providing it “to partners across the creative industries." After starting Stability in 2019, Mostaque built the company into an early AI juggernaut by seizing upon a promising research project that would become Stable Diffusion and funding it into a business reality. The ease with which the software generated detailed images from the simplest text prompts immediately captivated the public: 10 million people used it on any given day, the company told Forbes in early 2023. For some true believers, Mostaque was a crucial advocate for open-source AI development in a space dominated by the closed systems of OpenAI, Google and Anthropic. But his startup’s rise to one of the buzziest in generative AI was in part built on a series of exaggerations and misleading claims, as Forbes first reported last year (Mostaque disputed some points at the time). And they continued after he raised $100 million at a $1 billion valuation just days after launching Stable Diffusion in 2022. His failure to deliver on an array of grand promises, like building bespoke AI models for nation states, and his decision to pour tens of millions into research without a sustainable business plan, eroded Stability’s foundations and jeopardized its future. "He was just giving shit away,” one former employee told Forbes. “That man legitimately wanted to transform the world. He actually wanted to train AI models for kids in Malawi. Was it practical? Absolutely not." By October 2023, Stability would have less than $4 million left in the bank, according to an internal memo prepared for a board meeting and reviewed by Forbes. And mounting debt, including months of overdue Amazon Web Services payments, had already left it in the red. To avoid legal penalties for skipping Americans staff’s payroll, the document explained, the London-based startup was considering delaying tax payments to the U.K. government. It was Stability’s armada of GPUs, the wildly powerful and equally expensive chips undergirding AI, that were so taxing the company’s finances. Hosted by AWS, they had long been one of Mostaque’s bragging points; he often touted them as one of the world’s 10 largest supercomputers. They were responsible for helping Stability’s researchers build and maintain one of the top AI image generators, as well as break important new ground on generative audio, video and 3D models. “Undeniably, Stability has continued to ship a lot of models,” said one former employee. “They may not have profited off of it, but the broader ecosystem benefitted in a huge, huge way.” But the costs associated with so much compute were now threatening to sink the company. According to an internal October financial forecast seen by Forbes, Stability was on track to spend $99 million on compute in 2023. It noted as well that Stability was “underpaying AWS bills for July (by $1M)” and “not planning to pay AWS at the end of October for August usage ($7M).” Then there were the September and October bills, plus $1 million owed to Google Cloud and $600,000 to GPU cloud data center CoreWeave. (Amazon, Google and CoreWeave declined to comment.) With an additional $54 million allocated to wages and operating expenses, Stability’s total projected costs for 2023 were $153 million. But according to its October financial report, its projected revenue for the calendar year was just $11 million. Stability was on track to lose more money per month than it made in an entire year. The company’s dire financial position had thoroughly soured Stability’s current investors, including Coatue, which had invested tens of millions in the company during its $101 million funding round in 2022. In the middle of 2023, Mostaque agreed to an independent audit after Coatue raised a series of concerns, according to a source with direct knowledge of the matter. The outcome of the investigation is unclear. Coatue declined to comment. Within a week of an early October board meeting where Mostaque shared that financial forecast, Lightspeed Venture Partners, another major investor, sent a letter to the board urging them to sell the company. The distressing numbers had “severely undermined” the firm’s confidence in Mostaque’s ability to lead the company. “In particular, we are surprised and deeply concerned by a cash position just now disclosed to us that is inconsistent with prior discussions on this topic,” Lightspeed’s general counsel Brett Nissenberg wrote in the letter, a copy of which was viewed by Forbes. “Lightspeed believes that the company is not likely financeable on terms that would assure the company’s long term sound financial position.” (Lightspeed declined a request for comment.) The calls for a sale led Stability to quietly begin looking for a buyer. Bloomberg reported in November that Stability approached AI startups Cohere and Jasper to gauge their interest. Stability denied this, and Jasper CEO Timothy Young did the same when reached for comment by Forbes. A Cohere representative declined to comment. But one prominent AI company confirmed that Mostaque’s representatives had reached out to them to test the waters. Those talks did not advance because “the numbers didn’t add up,” this person, who declined to be named due to the confidential nature of the talks, told Forbes. Stability also tried to court Samsung as a buyer, going so far as to redecorate its office in advance of a planned meeting with the Korean electronics giant. (Samsung said that it invested in Stability in 2023 and that it does not comment on M&A discussions.) Coatue had been calling for Mostaque’s resignation for months, according to a source with direct knowledge. But it and other investors were unable to oust him because he was the company’s majority shareholder. When they tried a different tact by rallying other investors to offer him a juicy equity package to resign, Mostaque refused, said two sources. By October, Coatue and Lightspeed had had enough. Coatue left the board and Lightspeed resigned its observer seat. “Emad infuriated our initial investors so much it’s just making it impossible for us to raise more money under acceptable terms,” one current Stability executive told Forbes. The early months of 2024 saw Stability’s already precarious position eroding further still. Employees were quietly laid off. Three people in a position to know estimated that at least 10% of staff were cut. And cash reserves continued to dwindle. Mostaque mentioned a lifeline at the October board meeting: $95 million in tentative funding from new investors, pending due diligence. But in the end, only a fraction of it was wired, two sources say, much of it from Intel, which Forbes has learned invested $20 million, a fraction of what was reported. (Intel did not return a request for comment by publication time.) Two hours after Forbes broke the news of Mostaque’s plans to step down as CEO, Stability issued a press release confirming his resignation. Chief operating officer Wong and chief technology officer Laforte have taken over in the interim. Mostaque, who said on X that he still owns a majority of the company, also stepped down from the board, which has now initiated a search for a permanent CEO. There is a lot of work to be done to turn things around, and very little time in which to do it. Said the current Stability executive, “There’s still a possibility of a turnaround story, but the odds drop by the day.” In July of 2023, Mostaque still thought he could pull it off. Halfway through the month, he shared a fundraising plan with his lieutenants. It was wildly optimistic, detailing the raise of $500 million in cash and another $750 million in computing facilities from marquee investors like Nvidia, Google, Intel and the World Bank (Nvidia and Google declined comment. Intel did not respond. The World Bank said it did not invest in Stability). In a Slack message reviewed by Forbes, Mostaque said Google was “willing to move fast” and the round was “likely to be oversubscribed.” It wasn’t. Three people with direct knowledge of these fundraising efforts told Forbes that while there was some interest in Stability, talks often stalled when it came time to disclose financials. Two of them noted that earlier in the year, Mostaque had simply stopped engaging with VCs who asked for numbers. Only one firm invested around that time: actor Ashton Kutcher’s Sound Ventures, which invested $35 million in the form of a convertible SAFE note during the second quarter, according to an internal document. (Sound Ventures did not respond to a request for comment.) And though he’d managed to score a meeting with Nvidia and its CEO Jensen Huang, it ended in disaster, according to two sources. “Under Jensen's microscopic questions, Emad just fell apart,” a source in position to know told Forbes. Huang quickly concluded Stability wasn’t ready for an investment from Nvidia, the sources said. Mostaque told Forbes in an email that he had not met with Huang since 2022, except to say “hello and what’s up a few times after.” His July 2023 message references a plan to raise $150 million from Nvidia. (Nvidia declined to comment.) After a June Forbes investigation citing more than 30 sources revealed Mostaque’s history of misleading claims, Mostaque struggled to raise funding, a Stability investor told Forbes. (Mostaque disputed the story at the time and called it "coordinated lies" in his email this week to Forbes). Increasingly, investors scrutinized his assertions and pressed for data. And Young, now the CEO of Jasper, turned down a verbal offer to be Stability’s president after reading the article, according to a source with direct knowledge of the matter. The collapse of the talks aggravated the board and other executives, who had hoped Young would compensate for the sales and business management skills that Mostaque lacked, according to four people in a position to know. (Young declined to comment.) When Stability’s senior leadership convened in London for the CogX conference in September, the financing had still not closed. There, a group of executives confronted Mostaque asking questions about the company’s cash position and runway, according to three people with direct knowledge of the incident. They did not get the clarity they’d hoped for. By October, Mostaque had reduced his fundraising target by more than 80%. The months that followed saw a steady drumbeat of departures — general counsel Adam Avrunin, vice presidents Mike Melnicki, Ed Newton-Rex and Joe Penna, chief people officer Ozden Onder — culminating in the demoralizing March exit of Stable Diffusion’s primary developers Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz. Rombach, who led the team, had been angling to leave for months, two sources said, first threatening to resign last summer because of the fundraising failures. Others left over concerns about cash flow, as well as liabilities — including what four people described as Mostaque’s lax approach to ensuring that Stability products could not be used to produce child sexual abuse imagery. “Stability AI is committed to preventing the misuse of AI and prohibits the use of our image models and services for unlawful activity, including attempts to edit or create CSAM,” Ella Irwin, senior vice president of integrity, said in a statement. Newton-Rex told Forbes he resigned because he disagreed with Stability’s position that training AI on copyrighted work without consent is fair use. Melnicki and Penna declined to comment. Avrunin and Onder could not be reached for comment. None of the researchers responded to requests for comment. The Stable Diffusion researchers’ departure as a cohort says a lot about the state of Stability AI. The company’s researchers were widely viewed as its crown jewels, their work subsidized with a firehose of pricey compute power that was even extended to people outside the company. Martino Russi, an artificial intelligence researcher, told Forbes that though he was never formally employed by Stability, the company provided him a “staggering” amount of compute between January and April 2023 to play around with developing an AI video generator that Stability might someday use. “It was Candy Land or Coney Island,” said Russi, who estimates that his experiment, which was ultimately shelved, cost the company $2.5 million. Stable Diffusion was simultaneously Stability’s marquee product and its existential cash crisis. One current employee described it to Forbes as “a giant vacuum that absorbed everything: money, compute, people.” While the software was widely used, with Mostaque claiming downloads reaching into the hundreds of millions, Stability struggled to translate that wild success into revenue. Mostaque knew it could be done — peers at Databricks, Elastic and MongoDB had all turned a free product into a lucrative business — he just couldn’t figure out how. His first attempt was Stability’s API, which allowed paying customers to integrate Stable Diffusion into their own products. In early 2023, a handful of small companies, like art generator app NightCafe and presentation software startup Tome, signed on, according to four people with knowledge of the deals. But Stability’s poor account management services soured many, and in a matter of months NightCafe and Tome canceled their contracts, three people said. NightCafe founder Angus Russell told Forbes that his company switched to a competitor which “offered much cheaper inference costs and a broader service.” Tome did not respond to a request for comment. Meanwhile, Mostaque’s efforts to court larger companies like Samsung and Snapchat were failing, according to five people familiar with the effort. Canva, which was already one of the heaviest users of open-sourced Stable Diffusion, had multiple discussions with Stability, which was angling for a contract it hoped would generate several millions in annual revenue. But the deal never materialized, four sources said. “These three companies wanted and needed us,” one former employee told Forbes. “They would have been the perfect customers.” (Samsung, Snap and Canva declined to comment.) “It’s not that there was not an appetite to pay Stability — there were tons of companies that would have that wanted to,” the former employee said. “There was a huge opportunity and demand, but just a resistance to execution.” Mostaque’s other big idea was to provide governments with bespoke national AI models that would invigorate their economies and citizenry. “Emad envisions a world where AI through 100 national models serves not as a tool of the few, but as a benefactor to all promising to confront great adversaries, cancer, autism, and the sands of time itself,” the AI avatar of Aristotle said in his intro at the conference. Mostaque told several prospective customers that he could deliver such models within 60 days — an untenable timeline, according to two people in position to know. Stability attempted to develop a model for the Singaporean government over the protestation of employees who questioned its technical feasibility, three sources familiar with the effort told Forbes. But it couldn’t pull it off and Singapore never became a customer. (The government of Singapore confirmed it did not enter into a deal with Stability, but declined to answer additional questions.) As Stability careened from one new business idea to another, resources were abruptly reallocated and researchers reassigned. The whiplash shifts in a largely siloed organization demoralized and infuriated employees. “There were ‘urgent’ things, ‘urgent urgent’ things and ‘most urgent,’” one former employee complained. “None of these things seem important if everything is important.” Another former Stability executive was far more pointed in their assessment. “Emad is the most disorganized leader I have ever worked with in my career,” this person told Forbes. “He has no vision, and changes directions every week, often based on what he sees on Twitter.” In a video interview posted shortly before this story was published, Mostaque explained his leadership style: “I'm particularly great at taking creatives, developers, researchers, others, and achieving their full potential in designing systems. But I should not be dealing with, you know, HR and operations and business development and other elements. There are far better people than me to do that.” By December 2023, Stability had partially abandoned its open-source roots and announced that any commercial use of Stable Diffusion would cost customers at least $20 per month (non-commercial and research use of Stable Diffusion would remain free). But privately, Stability was considering a potentially more lucrative source of revenue: reselling the compute it was leasing from providers like AWS, according to six people familiar with the effort. Though it was essentially GPU arbitrage, Stability framed the strategy to investors as a “managed services” offering. Its damning October financial report projected optimistically that such an offering would bring in $139 million in 2024 — 98% of its revenue. Multiple employees at the time told Forbes they feared reselling compute, even if the company called it “managed services,” would violate the terms of Stability’s contract with AWS. Amazon declined to comment. “The line internally was that we are not reselling compute,” one former employee said. “This was some of the dirtiest feeling stuff.” Stability also discussed reselling a cluster of Nvidia A100 chips, leased via CoreWeave, to the venture capital firm Andreessen Horowitz, three sources said. “It was under the guise of managed services, but there wasn’t any management happening,” one of these people told Forbes. Andreessen Horowitz and CoreWeave declined to comment. Stability did not respond to questions about if it plans to continue this strategy now that Mostaque is out of the picture. Regardless, interim co-CEOs Wong and Laforte are on a tight timeline to clean up his mess. Board chairman Jim O’Shaughnessy said in a statement that he was confident the pair “will adeptly steer the company forward in developing and commercializing industry-leading generative AI products.” But burn continues to far outpace revenue. The Financial Times reported Friday that the company made $5.4 million of revenue in February, against $8 million in costs. Several sources said there are ongoing concerns about making payroll for the roughly 150 remaining employees. Leadership roles have gone vacant for months amid the disarray, leaving the company increasingly directionless. Meanwhile, a potentially catastrophic legal threat looms over the company: A trio of copyright infringement lawsuits brought by Getty Images and a group of artists in the U.S. and U.K., who claim Stability illegally used their art and photography to train the AI models powering Stable Diffusion. A London-based court has already rejected the company’s bid to throw out one of the lawsuits on the basis that none of its researchers were based in the U.K. And Stability’s claim that Getty’s Delaware lawsuit should be blocked because it's a U.K.-based company was rejected. (Stability did not respond to questions about the litigation.) AI-related copyright litigation “could go on for years,” according to Eric Goldman, a law professor at Santa Clara University. He told Forbes that though plaintiffs suing AI firms face an uphill battle overcoming the existing legal precedent on copyright infringement, the quantity of arguments available to make are virtually inexhaustible. “Like in military theory, if there’s a gap in your lines, that’s where the enemy pours through — if any one of those arguments succeeds, it could completely change the generative AI environment,” he said. “In some sense, generative AI as an industry has to win everything.” Stability, which had more than $100 million in the bank just a year and a half ago, is in a deep hole. Not only does it need more funding, it needs a viable business model — or a buyer with the vision and chops to make it successful in a fast-moving and highly competitive sector. At an all hands meeting this past Monday, Stability’s new leaders detailed a path forward. One point of emphasis: a plan to better manage resources and expenses, according to one person in attendance. It’s a start, but Mostaque’s meddling has left them with little runway to execute. His resignation, though, has given some employees hope. “A few people are 100% going to reconsider leaving after today,” said one current employee. “And the weird gloomy aura of hearing Emad talking nonsense for an hour is gone.” Shortly before Mostaque resigned, one current Stability executive told Forbes that they were optimistic his departure could make Stability appealing enough to receive a small investment or sale to a friendly party. “There are companies that have raised hundreds of millions of dollars that have much less intrinsic value than Stability,” the person said. “A white knight may still appear.”

#reddit #stability #ai’s

0

0.667

milaworld•This week

[N] How Stability AI’s Founder Tanked His Billion-Dollar Startup

forbes article: https://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup/ archive no paywall: https://archive.is/snbeV How Stability AI’s Founder Tanked His Billion-Dollar Startup Mar 29, 2024 Stability AI founder Emad Mostaque took the stage last week at the Terranea Resort in Palos Verdes, California to roaring applause and an introduction from an AI-generated Aristotle who announced him as “a modern Prometheus” with “the astuteness of Athena and the vision of Daedalus.” “Under his stewardship, AI becomes the Herculean force poised to vanquish the twin serpents of illness and ailment and extend the olive branch of longevity,” the faux Aristotle proclaimed. “I think that’s the best intro I’ve ever had,” Mostaque said. But behind Mostaque's hagiographic introduction lay a grim and fast metastasizing truth. Stability, once one of AI’s buzziest startups, was floundering. It had been running out of money for months and Mostaque had been unable to secure enough additional funding. It had defaulted on payments to Amazon whose cloud service undergirded Stability’s core offerings. The star research team behind its flagship text-to-image generator Stable Diffusion had tendered their resignations just three days before — as Forbes would first report — and other senior leaders had issued him an ultimatum: resign, or we walk too. Still, onstage before a massive audience of peers and acolytes, Mostaque talked a big game. “AI is jet planes for the mind,” he opined. “AI is our collective intelligence. It's the human Colossus.” He claimed a new, faster version of the Stable Diffusion image generator released earlier this month could generate “200 cats with hats per second.” But later, when he was asked about Stability’s financial model, Mostaque fumbled. “I can’t say that publicly,” he replied. “But it’s going well. We’re ahead of forecast.” Four days later, Mostaque stepped down as CEO of Stability, as Forbes first reported. In a post to X, the service formerly known as Twitter, he claimed he’d voluntarily abdicated his role to decentralize “the concentration of power in AI.” But sources told Forbes that was hardly the case. Behind the scenes, Mostaque had fought to maintain his position and control despite mounting pressure externally and internally to step down. Company documents and interviews with 32 current and former employees, investors, collaborators and industry observers suggest his abrupt exit was the result of poor business judgment and wild overspending that undermined confidence in his vision and leadership, and ultimately kneecapped the company. Mostaque, through his attorneys, declined to comment on record on a detailed list of questions about the reporting in this story. But in an email to Forbes earlier this week he broadly disputed the allegations. “Nobody tells you how hard it is to be a CEO and there are better CEOs than me to scale a business,” he said in a statement. “I am not sure anyone else would have been able to build and grow the research team to build the best and most widely used models out there and I’m very proud of the team there. I look forward to moving onto the next problem to handle and hopefully move the needle.” In an emailed statement, Christian Laforte and Shan Shan Wong, the interim co-CEOs who replaced Mostaque, said, "the company remains focused on commercializing its world leading technology” and providing it “to partners across the creative industries." After starting Stability in 2019, Mostaque built the company into an early AI juggernaut by seizing upon a promising research project that would become Stable Diffusion and funding it into a business reality. The ease with which the software generated detailed images from the simplest text prompts immediately captivated the public: 10 million people used it on any given day, the company told Forbes in early 2023. For some true believers, Mostaque was a crucial advocate for open-source AI development in a space dominated by the closed systems of OpenAI, Google and Anthropic. But his startup’s rise to one of the buzziest in generative AI was in part built on a series of exaggerations and misleading claims, as Forbes first reported last year (Mostaque disputed some points at the time). And they continued after he raised $100 million at a $1 billion valuation just days after launching Stable Diffusion in 2022. His failure to deliver on an array of grand promises, like building bespoke AI models for nation states, and his decision to pour tens of millions into research without a sustainable business plan, eroded Stability’s foundations and jeopardized its future. "He was just giving shit away,” one former employee told Forbes. “That man legitimately wanted to transform the world. He actually wanted to train AI models for kids in Malawi. Was it practical? Absolutely not." By October 2023, Stability would have less than $4 million left in the bank, according to an internal memo prepared for a board meeting and reviewed by Forbes. And mounting debt, including months of overdue Amazon Web Services payments, had already left it in the red. To avoid legal penalties for skipping Americans staff’s payroll, the document explained, the London-based startup was considering delaying tax payments to the U.K. government. It was Stability’s armada of GPUs, the wildly powerful and equally expensive chips undergirding AI, that were so taxing the company’s finances. Hosted by AWS, they had long been one of Mostaque’s bragging points; he often touted them as one of the world’s 10 largest supercomputers. They were responsible for helping Stability’s researchers build and maintain one of the top AI image generators, as well as break important new ground on generative audio, video and 3D models. “Undeniably, Stability has continued to ship a lot of models,” said one former employee. “They may not have profited off of it, but the broader ecosystem benefitted in a huge, huge way.” But the costs associated with so much compute were now threatening to sink the company. According to an internal October financial forecast seen by Forbes, Stability was on track to spend $99 million on compute in 2023. It noted as well that Stability was “underpaying AWS bills for July (by $1M)” and “not planning to pay AWS at the end of October for August usage ($7M).” Then there were the September and October bills, plus $1 million owed to Google Cloud and $600,000 to GPU cloud data center CoreWeave. (Amazon, Google and CoreWeave declined to comment.) With an additional $54 million allocated to wages and operating expenses, Stability’s total projected costs for 2023 were $153 million. But according to its October financial report, its projected revenue for the calendar year was just $11 million. Stability was on track to lose more money per month than it made in an entire year. The company’s dire financial position had thoroughly soured Stability’s current investors, including Coatue, which had invested tens of millions in the company during its $101 million funding round in 2022. In the middle of 2023, Mostaque agreed to an independent audit after Coatue raised a series of concerns, according to a source with direct knowledge of the matter. The outcome of the investigation is unclear. Coatue declined to comment. Within a week of an early October board meeting where Mostaque shared that financial forecast, Lightspeed Venture Partners, another major investor, sent a letter to the board urging them to sell the company. The distressing numbers had “severely undermined” the firm’s confidence in Mostaque’s ability to lead the company. “In particular, we are surprised and deeply concerned by a cash position just now disclosed to us that is inconsistent with prior discussions on this topic,” Lightspeed’s general counsel Brett Nissenberg wrote in the letter, a copy of which was viewed by Forbes. “Lightspeed believes that the company is not likely financeable on terms that would assure the company’s long term sound financial position.” (Lightspeed declined a request for comment.) The calls for a sale led Stability to quietly begin looking for a buyer. Bloomberg reported in November that Stability approached AI startups Cohere and Jasper to gauge their interest. Stability denied this, and Jasper CEO Timothy Young did the same when reached for comment by Forbes. A Cohere representative declined to comment. But one prominent AI company confirmed that Mostaque’s representatives had reached out to them to test the waters. Those talks did not advance because “the numbers didn’t add up,” this person, who declined to be named due to the confidential nature of the talks, told Forbes. Stability also tried to court Samsung as a buyer, going so far as to redecorate its office in advance of a planned meeting with the Korean electronics giant. (Samsung said that it invested in Stability in 2023 and that it does not comment on M&A discussions.) Coatue had been calling for Mostaque’s resignation for months, according to a source with direct knowledge. But it and other investors were unable to oust him because he was the company’s majority shareholder. When they tried a different tact by rallying other investors to offer him a juicy equity package to resign, Mostaque refused, said two sources. By October, Coatue and Lightspeed had had enough. Coatue left the board and Lightspeed resigned its observer seat. “Emad infuriated our initial investors so much it’s just making it impossible for us to raise more money under acceptable terms,” one current Stability executive told Forbes. The early months of 2024 saw Stability’s already precarious position eroding further still. Employees were quietly laid off. Three people in a position to know estimated that at least 10% of staff were cut. And cash reserves continued to dwindle. Mostaque mentioned a lifeline at the October board meeting: $95 million in tentative funding from new investors, pending due diligence. But in the end, only a fraction of it was wired, two sources say, much of it from Intel, which Forbes has learned invested $20 million, a fraction of what was reported. (Intel did not return a request for comment by publication time.) Two hours after Forbes broke the news of Mostaque’s plans to step down as CEO, Stability issued a press release confirming his resignation. Chief operating officer Wong and chief technology officer Laforte have taken over in the interim. Mostaque, who said on X that he still owns a majority of the company, also stepped down from the board, which has now initiated a search for a permanent CEO. There is a lot of work to be done to turn things around, and very little time in which to do it. Said the current Stability executive, “There’s still a possibility of a turnaround story, but the odds drop by the day.” In July of 2023, Mostaque still thought he could pull it off. Halfway through the month, he shared a fundraising plan with his lieutenants. It was wildly optimistic, detailing the raise of $500 million in cash and another $750 million in computing facilities from marquee investors like Nvidia, Google, Intel and the World Bank (Nvidia and Google declined comment. Intel did not respond. The World Bank said it did not invest in Stability). In a Slack message reviewed by Forbes, Mostaque said Google was “willing to move fast” and the round was “likely to be oversubscribed.” It wasn’t. Three people with direct knowledge of these fundraising efforts told Forbes that while there was some interest in Stability, talks often stalled when it came time to disclose financials. Two of them noted that earlier in the year, Mostaque had simply stopped engaging with VCs who asked for numbers. Only one firm invested around that time: actor Ashton Kutcher’s Sound Ventures, which invested $35 million in the form of a convertible SAFE note during the second quarter, according to an internal document. (Sound Ventures did not respond to a request for comment.) And though he’d managed to score a meeting with Nvidia and its CEO Jensen Huang, it ended in disaster, according to two sources. “Under Jensen's microscopic questions, Emad just fell apart,” a source in position to know told Forbes. Huang quickly concluded Stability wasn’t ready for an investment from Nvidia, the sources said. Mostaque told Forbes in an email that he had not met with Huang since 2022, except to say “hello and what’s up a few times after.” His July 2023 message references a plan to raise $150 million from Nvidia. (Nvidia declined to comment.) After a June Forbes investigation citing more than 30 sources revealed Mostaque’s history of misleading claims, Mostaque struggled to raise funding, a Stability investor told Forbes. (Mostaque disputed the story at the time and called it "coordinated lies" in his email this week to Forbes). Increasingly, investors scrutinized his assertions and pressed for data. And Young, now the CEO of Jasper, turned down a verbal offer to be Stability’s president after reading the article, according to a source with direct knowledge of the matter. The collapse of the talks aggravated the board and other executives, who had hoped Young would compensate for the sales and business management skills that Mostaque lacked, according to four people in a position to know. (Young declined to comment.) When Stability’s senior leadership convened in London for the CogX conference in September, the financing had still not closed. There, a group of executives confronted Mostaque asking questions about the company’s cash position and runway, according to three people with direct knowledge of the incident. They did not get the clarity they’d hoped for. By October, Mostaque had reduced his fundraising target by more than 80%. The months that followed saw a steady drumbeat of departures — general counsel Adam Avrunin, vice presidents Mike Melnicki, Ed Newton-Rex and Joe Penna, chief people officer Ozden Onder — culminating in the demoralizing March exit of Stable Diffusion’s primary developers Robin Rombach, Andreas Blattmann, Patrick Esser and Dominik Lorenz. Rombach, who led the team, had been angling to leave for months, two sources said, first threatening to resign last summer because of the fundraising failures. Others left over concerns about cash flow, as well as liabilities — including what four people described as Mostaque’s lax approach to ensuring that Stability products could not be used to produce child sexual abuse imagery. “Stability AI is committed to preventing the misuse of AI and prohibits the use of our image models and services for unlawful activity, including attempts to edit or create CSAM,” Ella Irwin, senior vice president of integrity, said in a statement. Newton-Rex told Forbes he resigned because he disagreed with Stability’s position that training AI on copyrighted work without consent is fair use. Melnicki and Penna declined to comment. Avrunin and Onder could not be reached for comment. None of the researchers responded to requests for comment. The Stable Diffusion researchers’ departure as a cohort says a lot about the state of Stability AI. The company’s researchers were widely viewed as its crown jewels, their work subsidized with a firehose of pricey compute power that was even extended to people outside the company. Martino Russi, an artificial intelligence researcher, told Forbes that though he was never formally employed by Stability, the company provided him a “staggering” amount of compute between January and April 2023 to play around with developing an AI video generator that Stability might someday use. “It was Candy Land or Coney Island,” said Russi, who estimates that his experiment, which was ultimately shelved, cost the company $2.5 million. Stable Diffusion was simultaneously Stability’s marquee product and its existential cash crisis. One current employee described it to Forbes as “a giant vacuum that absorbed everything: money, compute, people.” While the software was widely used, with Mostaque claiming downloads reaching into the hundreds of millions, Stability struggled to translate that wild success into revenue. Mostaque knew it could be done — peers at Databricks, Elastic and MongoDB had all turned a free product into a lucrative business — he just couldn’t figure out how. His first attempt was Stability’s API, which allowed paying customers to integrate Stable Diffusion into their own products. In early 2023, a handful of small companies, like art generator app NightCafe and presentation software startup Tome, signed on, according to four people with knowledge of the deals. But Stability’s poor account management services soured many, and in a matter of months NightCafe and Tome canceled their contracts, three people said. NightCafe founder Angus Russell told Forbes that his company switched to a competitor which “offered much cheaper inference costs and a broader service.” Tome did not respond to a request for comment. Meanwhile, Mostaque’s efforts to court larger companies like Samsung and Snapchat were failing, according to five people familiar with the effort. Canva, which was already one of the heaviest users of open-sourced Stable Diffusion, had multiple discussions with Stability, which was angling for a contract it hoped would generate several millions in annual revenue. But the deal never materialized, four sources said. “These three companies wanted and needed us,” one former employee told Forbes. “They would have been the perfect customers.” (Samsung, Snap and Canva declined to comment.) “It’s not that there was not an appetite to pay Stability — there were tons of companies that would have that wanted to,” the former employee said. “There was a huge opportunity and demand, but just a resistance to execution.” Mostaque’s other big idea was to provide governments with bespoke national AI models that would invigorate their economies and citizenry. “Emad envisions a world where AI through 100 national models serves not as a tool of the few, but as a benefactor to all promising to confront great adversaries, cancer, autism, and the sands of time itself,” the AI avatar of Aristotle said in his intro at the conference. Mostaque told several prospective customers that he could deliver such models within 60 days — an untenable timeline, according to two people in position to know. Stability attempted to develop a model for the Singaporean government over the protestation of employees who questioned its technical feasibility, three sources familiar with the effort told Forbes. But it couldn’t pull it off and Singapore never became a customer. (The government of Singapore confirmed it did not enter into a deal with Stability, but declined to answer additional questions.) As Stability careened from one new business idea to another, resources were abruptly reallocated and researchers reassigned. The whiplash shifts in a largely siloed organization demoralized and infuriated employees. “There were ‘urgent’ things, ‘urgent urgent’ things and ‘most urgent,’” one former employee complained. “None of these things seem important if everything is important.” Another former Stability executive was far more pointed in their assessment. “Emad is the most disorganized leader I have ever worked with in my career,” this person told Forbes. “He has no vision, and changes directions every week, often based on what he sees on Twitter.” In a video interview posted shortly before this story was published, Mostaque explained his leadership style: “I'm particularly great at taking creatives, developers, researchers, others, and achieving their full potential in designing systems. But I should not be dealing with, you know, HR and operations and business development and other elements. There are far better people than me to do that.” By December 2023, Stability had partially abandoned its open-source roots and announced that any commercial use of Stable Diffusion would cost customers at least $20 per month (non-commercial and research use of Stable Diffusion would remain free). But privately, Stability was considering a potentially more lucrative source of revenue: reselling the compute it was leasing from providers like AWS, according to six people familiar with the effort. Though it was essentially GPU arbitrage, Stability framed the strategy to investors as a “managed services” offering. Its damning October financial report projected optimistically that such an offering would bring in $139 million in 2024 — 98% of its revenue. Multiple employees at the time told Forbes they feared reselling compute, even if the company called it “managed services,” would violate the terms of Stability’s contract with AWS. Amazon declined to comment. “The line internally was that we are not reselling compute,” one former employee said. “This was some of the dirtiest feeling stuff.” Stability also discussed reselling a cluster of Nvidia A100 chips, leased via CoreWeave, to the venture capital firm Andreessen Horowitz, three sources said. “It was under the guise of managed services, but there wasn’t any management happening,” one of these people told Forbes. Andreessen Horowitz and CoreWeave declined to comment. Stability did not respond to questions about if it plans to continue this strategy now that Mostaque is out of the picture. Regardless, interim co-CEOs Wong and Laforte are on a tight timeline to clean up his mess. Board chairman Jim O’Shaughnessy said in a statement that he was confident the pair “will adeptly steer the company forward in developing and commercializing industry-leading generative AI products.” But burn continues to far outpace revenue. The Financial Times reported Friday that the company made $5.4 million of revenue in February, against $8 million in costs. Several sources said there are ongoing concerns about making payroll for the roughly 150 remaining employees. Leadership roles have gone vacant for months amid the disarray, leaving the company increasingly directionless. Meanwhile, a potentially catastrophic legal threat looms over the company: A trio of copyright infringement lawsuits brought by Getty Images and a group of artists in the U.S. and U.K., who claim Stability illegally used their art and photography to train the AI models powering Stable Diffusion. A London-based court has already rejected the company’s bid to throw out one of the lawsuits on the basis that none of its researchers were based in the U.K. And Stability’s claim that Getty’s Delaware lawsuit should be blocked because it's a U.K.-based company was rejected. (Stability did not respond to questions about the litigation.) AI-related copyright litigation “could go on for years,” according to Eric Goldman, a law professor at Santa Clara University. He told Forbes that though plaintiffs suing AI firms face an uphill battle overcoming the existing legal precedent on copyright infringement, the quantity of arguments available to make are virtually inexhaustible. “Like in military theory, if there’s a gap in your lines, that’s where the enemy pours through — if any one of those arguments succeeds, it could completely change the generative AI environment,” he said. “In some sense, generative AI as an industry has to win everything.” Stability, which had more than $100 million in the bank just a year and a half ago, is in a deep hole. Not only does it need more funding, it needs a viable business model — or a buyer with the vision and chops to make it successful in a fast-moving and highly competitive sector. At an all hands meeting this past Monday, Stability’s new leaders detailed a path forward. One point of emphasis: a plan to better manage resources and expenses, according to one person in attendance. It’s a start, but Mostaque’s meddling has left them with little runway to execute. His resignation, though, has given some employees hope. “A few people are 100% going to reconsider leaving after today,” said one current employee. “And the weird gloomy aura of hearing Emad talking nonsense for an hour is gone.” Shortly before Mostaque resigned, one current Stability executive told Forbes that they were optimistic his departure could make Stability appealing enough to receive a small investment or sale to a friendly party. “There are companies that have raised hundreds of millions of dollars that have much less intrinsic value than Stability,” the person said. “A white knight may still appear.”

#reddit #stability #ai’s

10y of product development, 2 bankruptcies, and 1 Exit — what next? [Extended Story]

0

1

Slight-Explanation29•This week

10y of product development, 2 bankruptcies, and 1 Exit — what next? [Extended Story]

10 years of obsessive pursuit from the bottom to impressive product-market fit and exit. Bootstrapping tech products as Software Developer and 3x Startup Founder (2 bankruptcies and 1 exit). Hi everyone, your motivation has inspired me to delve deeper into my story. So, as promised to some of you, I've expanded on it a bit more, along with my brief reflections. There are many founders, product creators, and proactive individuals, I’ve read many of your crazy stories and lessons so I decided to share mine and the lessons I learned from the bottom to impressive product-market fit and exit. I've spent almost the past 10 years building tech products as a Corporate Team Leader, Senior Software Developer, Online Course Creator, Programming Tutor, Head of Development/CTO, and 3x Startup Founder (2 bankruptcies, and 1 exit). And what next? good question... A brief summary of my journey: Chapter 1: Software Developer / Team Leader / Senior Software Developer I’ve always wanted to create products that win over users’ hearts, carry value, and influence users. Ever since my school days, I’ve loved the tech part of building digital products. At the beginning of school, I started hosting servers for games, blogs and internet forums, and other things that did not require much programming knowledge. My classmates and later even over 100 people played on servers that I hosted on my home PC. Later, as the only person in school, I passed the final exam in computer science. During my computer science studies, I started my first job as a software developer. It was crazy, I was spending 200–300 hours a month in the office attending also to daily classes. Yes, I didn’t have a life, but it truly was the fulfillment of my dreams. I was able to earn good money doing what I love, and I devoted fully myself to it. My key to effectively studying IT and growing my knowledge at rocket speed was learning day by day reading guides, building products to the portfolio, watching youtube channels and attending conferences, and even watching them online, even if I didn’t understand everything at the beginning. In one year we’ve been to every possible event within 400km. We were building healthcare products that were actually used in hospitals and medical facilities. It was a beautiful adventure and tons of knowledge I took from this place. That time I built my first product teams, hired many great people, and over the years became a senior developer and team leader. Even I convinced my study mates to apply to this company and we studied together and worked as well. Finally, there were 4 of us, when I left a friend of mine took over my position and still works there. If you’re reading this, I’m sending you a flood of love and appreciation. I joined as the 8th person, and after around 4 years, when I left hungry for change, there were already over 30 of us, now around 100. It was a good time, greetings to everyone. I finished my Master’s and Engineering degrees in Computer Science, and it was time for changes. Chapter 2: 1st time as a Co-founder — Marketplace In the meantime, there was also my first startup (a marketplace) with four of my friends. We all worked on the product, each of us spent thousands of hours, after hours, entire weekends… and I think finally over a year of work. As you might guess, we lacked the most important things: sales, marketing, and product-market fit. We thought users think like us. We all also worked commercially, so the work went very smoothly, but we didn’t know what we should do next with it… Finally, we didn’t have any customers, but you know what, I don’t regret it, a lot of learning things which I used many times later. The first attempts at validating the idea with the market and business activities. In the end, the product was Airbnb-sized. Landing pages, listings, user panels, customer panels, admin site, notifications, caches, queues, load balancing, and much more. We wanted to publish the fully ready product to the market. It was a marketplace, so if you can guess, we had to attract both sides to be valuable. “Marketplace” — You can imagine something like Uber, if you don’t have passengers it was difficult to convince taxi drivers, if you don’t have a large number of taxi drivers you cannot attract passengers. After a year of development, we were overloaded, and without business, marketing, sales knowledge, and budget. Chapter 3: Corp Team Lead / Programming Tutor / Programming Architecture Workshop Leader Working in a corporation, a totally different environment, an international fintech, another learning experience, large products, and workmates who were waiting for 5 pm to finish — it wasn’t for me. Very slow product development, huge hierarchy, being an ant at the bottom, and low impact on the final product. At that time I understood that being a software developer is not anything special and I compared my work to factory worker. Sorry for that. High rates have been pumped only by high demand. Friends of mine from another industry do more difficult things and have a bigger responsibility for lower rates. That’s how the market works. This lower responsibility time allowed for building the first online course after hours, my own course platform, individual teaching newbies programming, and my first huge success — my first B2C customers, and B2B clients for workshops. I pivoted to full focus on sales, marketing, funnels, advertisements, demand, understanding the market, etc. It was 10x easier than startups but allowed me to learn and validate my conceptions and ideas on an easier market and showed me that it’s much easier to locate their problem/need/want and create a service/product that responds to it than to convince people of your innovative ideas. It’s just supply and demand, such a simple and basic statement, in reality, is very deep and difficult to understand without personal experience. If you’re inexperienced and you think you understand, you don’t. To this day, I love to analyze this catchword in relation to various industries / services / products and rediscover it again and again... While writing this sentence, I’m wondering if I’m not obsessed. Chapter 4: Next try — 2nd time as a founder — Edtech Drawing upon my experiences in selling services, offering trainings, and teaching programming, I wanted to broaden my horizons, delve into various fields of knowledge, involve more teachers, and so on. We started with simple services in different fields of knowledge, mainly relying on teaching in the local area (without online lessons). As I had already gathered some knowledge and experience in marketing and sales, things were going well and were moving in the right direction. The number of teachers in various fields was growing, as was the number of students. I don’t remember the exact statistics anymore, but it was another significant achievement that brought me a lot of satisfaction and new experiences. As you know, I’m a technology lover and couldn’t bear to look at manual processes — I wanted to automate everything: lessons, payments, invoices, customer service, etc. That’s when I hired our first developers (if you’re reading this, I’m sending you a flood of love — we spent a lot of time together and I remember it as a very fruitful and great year) and we began the process of tool and automation development. After a year we had really extended tools for students, teachers, franchise owners, etc. We had really big goals, we wanted to climb higher and higher. Maybe I wouldn’t even fully call it Startup, as the client was paying for the lessons, not for the software. But it gave us positive income, bootstrap financing, and tool development for services provided. Scaling this model was not as costless as SaaS because customer satisfaction was mainly on the side of the teacher, not the quality of the product (software). Finally, we grew to nearly 10 people and dozens of teachers, with zero external funding, and almost $50k monthly revenue. We worked very hard, day and night, and by November 2019, we were packed with clients to the brim. And as you know, that’s when the pandemic hit. It turned everything upside down by 180 degrees. Probably no one was ready for it. With a drastic drop in revenues, society started to save. Tired from the previous months, we had to work even harder. We had to reduce the team, change the model, and save what we had built. We stopped the tool’s development and sales, and with the developers, we started supporting other product teams to not fire them in difficult times. The tool worked passively for the next two years, reducing incomes month by month. With a smaller team providing programming services, we had full stability and earned more than relying only on educational services. At the peak of the pandemic, I promised myself that it was the last digital product I built… Never say never… Chapter 5: Time for fintech — Senior Software Developer / Team Lead / Head of Development I worked for small startups and companies. Building products from scratch, having a significant impact on the product, and complete fulfillment. Thousands of hours and sacrifices. This article mainly talks about startups that I built, so I don’t want to list all the companies, products, and applications that I supported as a technology consultant. These were mainly start-ups with a couple of people up to around 100 people on board. Some of the products were just a rescue mission, others were building an entire tech team. I was fully involved in all of them with the hope that we would work together for a long time, but I wasn’t the only one who made mistakes when looking for a product-market fit. One thing I fully understood: You can’t spend 8–15 hours a day writing code, managing a tech team, and still be able to help build an audience. In marketing and sales, you need to be rested and very creative to bring results and achieve further results and goals. If you have too many responsibilities related to technology, it becomes ineffective. I noticed that when I have more free time, more time to think, and more time to bounce the ball against the wall, I come up with really working marketing/sales strategies and solutions. It’s impossible when you are focused on code all day. You must know that this chapter of my life was long and has continued until now. Chapter 6: 3rd time as a founder — sold Never say never… right?\\ It was a time when the crypto market was really high and it was really trending topic. You know that I love technology right? So I cannot miss the blockchain world. I had experience in blockchain topics by learning on my own and from startups where I worked before. I was involved in crypto communities and I noticed a “starving crowd”. People who did things manually and earned money(crypto) on it.I found potential for building a small product that solves a technological problem. I said a few years before that I don’t want to start from scratch. I decided to share my observations and possibilities with my good friend. He said, “If you gonna built it, I’m in”. I couldn’t stop thinking about it. I had thought and planned every aspect of marketing and sales. And you know what. On this huge mindmap “product” was only one block. 90% of the mindmap was focused on marketing and sales. Now, writing this article, I understood what path I went from my first startup to this one. In the first (described earlier) 90% was the product, but in the last one 90% was sales and marketing. Many years later, I did this approach automatically. What has changed in my head over the years and so many mistakes? At that time, the company for which I provided services was acquired. The next day I got a thank you for my hard work and all my accounts were blocked. Life… I was shocked. We were simply replaced by their trusted technology managers. They wanted to get full control. They acted a bit unkindly, but I knew that they had all my knowledge about the product in the documentation, because I’m used to drawing everything so that in the moment of my weakness (illness, whatever) the team could handle it. That’s what solid leaders do, right? After a time, I know that these are normal procedures in financial companies, the point is that under the influence of emotions, do not do anything inappropriate. I quickly forgot about it, that I was brutally fired. All that mattered was to bring my plan to life. And it has been started, 15–20 hours a day every day. You have to believe me, getting back into the game was incredibly satisfying for me. I didn’t even know that I would be so excited. Then we also noticed that someone was starting to think about the same product as me. So the race began a game against time and the market. I assume that if you have reached this point, you are interested in product-market fit, marketing, and sales, so let me explain my assumptions to you: Product: A very very small tool that allowed you to automate proper tracking and creation of on-chain transactions. Literally, the whole app for the user was located on only three subpages. Starving Crowd: We tapped into an underserved market. The crypto market primarily operates via communities on platforms like Discord, Reddit, Twitter, Telegram, and so on. Therefore, our main strategy was directly communicating with users and demonstrating our tool. This was essentially “free marketing” (excluding the time we invested), as we did not need to invest in ads, promotional materials, or convince people about the efficacy of our tool. The community could directly observe on-chain transactions executed by our algorithms, which were processed at an exceptionally fast rate. This was something they couldn’t accomplish manually, so whenever someone conducted transactions using our algorithm, it was immediately noticeable and stirred a curiosity within the community (how did they do that!). Tests: I conducted the initial tests of the application on myself — we had already invested significantly in developing the product, but I preferred risking my own resources over that of the users. I provided the tool access to my wallet, containing 0.3ETH, and went to sleep. Upon waking up, I discovered that the transactions were successful and my wallet had grown to 0.99ETH. My excitement knew no bounds, it felt like a windfall. But, of course, there was a fair chance I could have lost it too. It worked. As we progressed, some users achieved higher results, but it largely hinged on the parameters set by them. As you can surmise, the strategy was simple — buy low, sell high. There was considerable risk involved. Churn: For those versed in marketing, the significance of repeat visitors cannot be overstated. Access to our tool was granted only after email verification and a special technique that I’d prefer to keep confidential. And this was all provided for free. While we had zero followers on social media, we saw an explosion in our email subscriber base and amassed a substantial number of users and advocates. Revenue Generation: Our product quickly gained popularity as we were effectively helping users earn — an undeniable value proposition. Now, it was time to capitalize on our efforts. We introduced a subscription model charging $300 per week or $1,000 per month — seemingly high rates, but the demand was so intense that it wasn’t an issue. Being a subscriber meant you were prioritized in the queue, ensuring you were among the first to reap benefits — thus adding more “value”. Marketing: The quality of our product and its ability to continually engage users contributed to it achieving what can best be described as viral. It was both a source of pride and astonishment to witness users sharing charts and analyses derived from our tool in forum discussions. They weren’t actively promoting our product but rather using screenshots from our application to illustrate certain aspects of the crypto world. By that stage, we had already assembled a team to assist with marketing, and programming, and to provide round-the-clock helpdesk support. Unforgettable Time: Despite the hype, my focus remained steadfast on monitoring our servers, their capacity, and speed. Considering we had only been on the market for a few weeks, we were yet to implement alerts, server scaling, etc. Our active user base spanned from Japan to the West Coast of the United States. Primarily, our application was used daily during the evenings, but considering the variety of time zones, the only time I could afford to sleep was during the evening hours in Far Eastern Europe, where we had the least users. However, someone always needed to be on guard, and as such, my phone was constantly by my side. After all, we couldn’t afford to let our users down. We found ourselves working 20 hours a day, catering to thousands of users, enduring physical fatigue, engaging in talks with VCs, and participating in conferences. Sudden Downturn: Our pinnacle was abruptly interrupted by the war in Ukraine (next macroeconomic shot straight in the face, lucky guy), a precipitous drop in cryptocurrency value, and swiftly emerging competition. By this time, there were 5–8 comparable tools had infiltrated the market. It was a challenging period as we continually stumbled upon new rivals. They immediately embarked on swift fundraising endeavors — a strategy we overlooked, which in retrospect was a mistake. Although our product was superior, the competitors’ rapid advancement and our insufficient funds for expeditious scaling posed significant challenges. Nonetheless, we made a good decision. We sold the product (exit) to competitors. The revenue from “exit” compensated for all the losses, leaving us with enough rest. We were a small team without substantial budgets for rapid development, and the risk of forming new teams without money to survive for more than 1–2 months was irresponsible. You have to believe me that this decision consumed us sleepless nights. Finally, we sold it. They turned off our app but took algorithms and users. Whether you believe it or not, after several months of toiling day and night, experiencing burnout, growing weary of the topic, and gaining an extra 15 kg in weight, we finally found our freedom… The exit wasn’t incredibly profitable, but we knew they had outdone us. The exit covered all our expenses and granted us a well-deserved rest for the subsequent quarter. It was an insane ride. Despite the uncertainty, stress, struggles, and sleepless nights, the story and experience will remain etched in my memory for the rest of my life. Swift Takeaways: Comprehending User Needs: Do you fully understand the product-market fit? Is your offering just an accessory or does it truly satisfy the user’s needs? The Power of Viral Marketing: Take inspiration from giants like Snapchat, ChatGPT, and Clubhouse. While your product might not attain the same scale (but remember, never say never…), the closer your concept is to theirs, the easier your journey will be. If your user is motivated to text a friend saying, “Hey, check out how cool this is” (like sharing ChatGPT), then you’re on the best track. Really. Even if it doesn’t seem immediately evident, there could be a way to incorporate this into your product. Keep looking until you find it. Niche targeting — the more specific and tailored your product is to a certain audience, the easier your journey will be People love buying from people — establishing a personal brand and associating yourself with the product can make things easier. Value: Seek to understand why users engage with your product and keep returning. The more specific and critical the issue you’re aiming to solve, the easier your path will be. Consider your offerings in terms of products and services and focus on sales and marketing, regardless of personal sentiments. These are just a few points, I plan to elaborate on all of them in a separate article. Many products undergo years of development in search of market fit, refining the user experience, and more. And guess what? There’s absolutely nothing wrong with that. Each product and market follows its own rules. Many startups have extensive histories before they finally make their mark (for instance, OpenAI). This entire journey spanned maybe 6–8 months. I grasped and capitalized on the opportunity, but we understood from the start that establishing a startup carried a significant risk, and our crypto product was 10 times riskier. Was it worth it? Given my passion for product development — absolutely. Was it profitable? — No, considering the hours spent — we lose. Did it provide a stable, problem-free life — nope. Did this entire adventure offer a wealth of happiness, joy, and unforgettable experiences — definitely yes. One thing is certain — we’ve amassed substantial experience and it’s not over yet :) So, what lies ahead? Chapter 7: Reverting to the contractor, developing a product for a crypto StartupReturning to the past, we continue our journey… I had invested substantial time and passion into the tech rescue mission product. I came on board as the technical Team Leader of a startup that had garnered over $20M in seed round funding, affiliated with the realm of cryptocurrencies. The investors were individuals with extensive backgrounds in the crypto world. My role was primarily technical, and there was an abundance of work to tackle. I was fully immersed, and genuinely devoted to the role. I was striving for excellence, knowing that if we secured another round of financing, the startup would accelerate rapidly. As for the product and marketing, I was more of an observer. After all, there were marketing professionals with decades of experience on board. These were individuals recruited from large crypto-related firms. I had faith in them, kept an eye on their actions, and focused on my own responsibilities. However, the reality was far from satisfactory. On the last day, the principal investor for the Series A round withdrew. The board made the tough decision to shut down. It was a period of intense observation and gaining experience in product management. This was a very brief summary of the last 10 years. And what next? (Last) Chapter 8: To be announced — Product Owner / Product Consultant / Strategist / CTO After spending countless hours and days deliberating my next steps, one thing is clear: My aspiration is to continue traversing the path of software product development, with the hopeful anticipation that one day, I might ride the crest of the next big wave and ascend to the prestigious status of a unicorn company. I find myself drawn to the process of building products, exploring product-market fit, strategizing, engaging in software development, seeking out new opportunities, networking, attending conferences, and continuously challenging myself by understanding the market and its competitive landscape. Product Owner / Product Consultant / CTO / COO: I’m not entirely sure how to categorize this role, as I anticipate that it will largely depend on the product to which I will commit myself fully. My idea is to find one startup/company that wants to build a product / or already has a product, want to speed up, or simply doesn’t know what’s next. Alternatively, I could be a part of an established company with a rich business history, which intends to invest in digitization and technological advancements. The goal would be to enrich their customer experience by offering complementary digital products Rather than initiating a new venture from ground zero with the same team, I am receptive to new challenges. I am confident that my past experiences will prove highly beneficial for the founders of promising, burgeoning startups that already possess a product, or are in the initial phases of development. ‘Consultant’ — I reckon we interpret this term differently. My aim is to be completely absorbed in a single product, crafting funnels, niches, strategies, and all that is necessary to repeatedly achieve the ‘product-market fit’ and significant revenue. To me, ‘consultant’ resonates more akin to freelancing than being an employee. My current goal is to kickstart as a consultant and aide, dealing with facilitating startups in their journey from point A to B. Here are two theoretical scenarios to illustrate my approach: Scenario 1: (Starting from point A) You have a product but struggle with marketing, adoption, software, strategy, sales, fundraising, or something else. I conduct an analysis and develop a strategy to reach point B. I take on the “dirty work” and implement necessary changes, including potential pivots or shifts (going all-in) to guide the product to point B. The goal is to reach point B, which could involve achieving a higher valuation, expanding the user base, increasing sales, or generating monthly revenue, among other metrics. Scenario 2: (Starting from point A) You have a plan or idea but face challenges with marketing, adoption, strategy, software, sales, fundraising, or something else. I analyze the situation and devise a strategy to reach point B. I tackle the necessary tasks, build the team, and overcome obstacles to propel the product to point B. I have come across the view that finding the elusive product-market fit is the job of the founder, and it’s hard for me to disagree. However, I believe that my support and experiences can help save money, many failures, and most importantly, time. I have spent a great deal of time learning from my mistakes, enduring failure after failure, and even had no one to ask for support or opinion, which is why I offer my help. Saving even a couple of years, realistically speaking, seems like a value I’m eager to provide… I invite you to share your thoughts and insights on these scenarios :) Closing Remarks: I appreciate your time and effort in reaching this point. This has been my journey, and I wouldn’t change it for the world. I had an extraordinary adventure, and now I’m ready for the next exciting battle with the market and new software products. While my entire narrative is centered around startups, especially the ones I personally built, I’m planning to share more insights drawn from all of my experiences, not just those as a co-founder. If you’re currently developing your product or even just considering the idea, I urge you to reach out to me. Perhaps together, we can create something monumental :) Thank you for your time and insights. I eagerly look forward to engaging in discussions and hearing your viewpoints. Please remember to like and subscribe. Nothing motivates to write more than positive feedback :) Matt.

#reddit #product #development,

Stable Video 3D creates orbiting videos of objects

0

1

yuki_taylor•This week

Stable Video 3D creates orbiting videos of objects

The introduction of Stable Video 3D (SV3D) by Stability AI could significantly change how we interact with digital content in several straightforward ways: \- Online Shopping: Products can be viewed in 360-degree videos, making shopping online more like seeing the item in person. This could help shoppers make better decisions and feel more confident about their purchases. \- Design and Gaming: Designers and game developers can quickly turn images into 3D models, speeding up their work and improving the quality of their designs and games. \- Augmented and Virtual Reality: Creating content for AR and VR could become easier and more accessible, leading to more immersive and engaging experiences in these platforms. \- Accessibility for Creators: With SV3D available for non-commercial use, more people can experiment with 3D content creation, potentially leading to innovative applications and uses in various fields. This literally just happened if you want AI news as it drops it launches here first. The whole article has been extrapolated here as well for convenience.

#reddit #stable #video

40% Of SMBs Still Can't Pay Their Rent, Extending High Delinquency From September Into October

0

1

Aegidius25•This week

40% Of SMBs Still Can't Pay Their Rent, Extending High Delinquency From September Into October

https://www.alignable.com/forum/q4s-off-to-a-rough-start-40-of-smbs-still-cant-pay-their-rent October 31, 2023: While the federal government reported a surge in economic growth for the U.S. last week, that news doesn't hold true for many small business owners. In fact, in October polling by Alignable, only 12% said their companies are experiencing significant growth this month. Beyond that, Alignable’s October Rent Report, released today, shows that a whopping 40% of SMBs couldn't even pay their October rent in full and on time. This marks the second consecutive month of a 40% rent delinquency rate -- extending 2023's record high from September through October. These findings are based on responses from 4,246 randomly selected small business owners surveyed from 10/1/23 to 10/30/23, as well as input from 44,000+ other respondents over the past year. As the chart below shows, October's SMB rent delinquency rate is 10 percentage points higher than it was in January, reflecting cumulative economic struggles: increased rents, high interest rates, still-stifling inflation, rising labor costs, and revenues that have declined since this time last year. Rent delinquency rates among small businesses during 2023 based on Alignable surveys So, Why's Rent Delinquency At 40% For A 2nd Month? Here’s the current list of problems contributing to two months' worth of the highest delinquency rate 2023 has seen so far: Consumer Spending Declines On Main Street: Quarterly, we ask about customer spending habits at retailers. This month, 45% of independent Mom and Pop Shops said spending has been down over the last 30 days. Some said it was due to more people spending money online with big retailers like Amazon. This figure is quite high, especially considering that back in July, only 24% reported a drop in consumer spending -- 21 percentage points less severe than it is now. Revenue Troubles: 42% are making half or less of the income they generated monthly prior to COVID. For businesses that are less than three years old, this situation is even worse: 53% of this group reports making half or less of what they generated this time last year. High Interest Rates: Over half of all SMB owners polled said the past 19 months of high interest rates have hurt their margins, reduced revenues, and put their expansion plans on hold, as they don't want to apply for loans. Increased Rent Prices: 50% say they’re being charged more for rent now than they were six months ago, with 15% saying rent has increased by 20% or more. At present, only 37% of pre-COVID businesses have recovered financially from the pandemic era, leaving 63% still striving to make up for time they lost due to COVID, inflationary pressures, and high interest rates. There's a slight silver lining here, though, as the 37% figure is three percentage points higher than it was in September. But, with that said, a recovery rate of 37% after more than three and a half years is still very low and speaks volumes about the ongoing list of troubles small business owners face looking into the rest of 2023. Tech, Manufacturing, Gyms, Beauty & Retail Struggle Examining the rent delinquency landscape in terms of sectors, there's quite a negative shift occurring among some industries in October. Let's look at the charts below to see what's really happening. Sectors most affected by rent delinquency include tech and retail Details on sectors affected by rent delinquency in October This is alarming for a few reasons: The countless technology layoffs at larger companies over the past year appear to be affecting the small companies now, too, who are often dependent on the larger ones as clients. Right now, 54% of science/technology small companies couldn't pay their October rent, up 10 percentage points from September and 16 percentage points since August. There are also some comments in the surveys of technology roles being reduced or replaced by ChatGPT and other AI, which can write software programs. Gyms have been struggling now for a while and now 50% of them can't afford the rent, up 8 percentage points from September. The biggest shift between October and September occurred among manufacturers, partially due to ongoing fluctuation in the price of gas and other inflationary issues. For quite some time, manufacturers were improving a lot in terms of their rent delinquency rates, but in October, they jumped 25 percentage points, doubling their rate, which is now 50%. This is also a record high for manufacturers in 2023. We hope this is just a blip, but we'll see in November. Also due, in part, to fluctuating gas prices and costs of vehicles, 45% of transportation companies couldn't pay October rent in full and on time. That's up 6 percentage points from last month. Sadly, 47% of salon owners couldn't cover October rent, after showing a lot of stability over the past few months. But that stability ended this month, as salons' rent delinquency rates jumped nine percentage points. Though rates have dropped three percentage points in October, a high percentage of retailers are still having trouble paying the rent. Last month, it was 47%. This month, it's better, but is still over 40%, landing at 44%. This is worrisome, especially since Q4 is a "make it or break it" time for many Main Street merchants. Looking more closely at the industries, there was some good news, in that a few others experienced lower delinquency rates in October, including restaurants, which dipped to 40% from 44% in September. Travel/lodging dropped seven percentage points to 38% (from 45% last month), as did education, which is also at 38%, down from 43%. When looking at rent delinquency from the vantage point of the states that are most affected, many surges can be seen between October and September, while a few states saw some dramatic, encouraging declines, too. Rent Troubles Increase For IL, VA, TX, MA, FL, & CO Looking at the states' charts, you can see how tumultuous the rent story has become this fall. Let's first talk about those with significant jumps in their delinquency rates. Here's the rundown: Illinois leads the list once again. After having a better month in September, its delinquency rate has soared, once more, landing at 54% for October (up from 46% last month). In fact, the 54% figure is the highest rate IL-based SMBs have seen in 2023. Virginia was in great shape last month, with a delinquency rate of just 19%. But Virginia-based small business owners have had a very rough month, at least in terms of rent. Now, 50% of them who took our poll say they couldn't cover rent (an increase of 31 percentage points). Texas is third on the list, with an 11-percentage-point lift from 38% in September to 49% in October. MA is next up at 48%, which marks the largest jump on the chart -- 32 percentage points from a low of just 16% in September. Small businesses in Florida have also experienced two challenging months in terms of rent delinquency. Right now, 45% of SMBs there couldn't afford to pay, up nine percentage points from September and 15 percentage points from August. Colorado's businesses regressed in October, hitting a new record high of 40%. That rent delinquency rate jumped 13 percentage points from September to October. While we just covered states with some very high delinquency rates, there were also several more positive swings that have occurred in October. Though encouraging, we'll have to see how long those delinquency rates continue. Here are the most remarkable: New York -- After reaching a record rate of 55% last month, New York's small business owners now report a more stable number: just 29%. That's down 26 percentage points. New Jersey -- New York's neighbor has an even more impressive story in October: only 20% of New Jersey's SMBs couldn't pay rent this month, a record low over at least the past 14 months, down 34 percentage points from a record high of 54%. Michigan -- Similarly, Michigan's small business owners boast a rate of just 20%, down from 45% in September.

#reddit #smbs #still

How is ML/AI implemented in your company?

0

1

mark1x12110•This week

How is ML/AI implemented in your company?

Hi, I am currently working as a full time SWE. I generally get to work with data/build solutions using data but I don't have any official ML/Data science title or background. Ideally, I'd like to be able to start making more intelligent decisions using that data and that's why I am interested on machine learning. Ideally, I'd like to write production systems that leverage models that I can build and understand end to end but after some research, I am not sure if such jobs exist today or if teams are usually split between modelling team (people that only train and analyze models) and SWE teams (people that take such models and focus on deploying them on production). My current company is using that model of development, so I might need to leave to be able to do what I want. I imagine startups work that way, but I am looking for some stability. Thanks!

#reddit #ml/ai #implemented

MMML | Deploy HuggingFace training model rapidly based on MetaSpore

0

1

qazmkopp•This week

MMML | Deploy HuggingFace training model rapidly based on MetaSpore

A few days ago, HuggingFace announced a $100 million Series C funding round, which was big news in open source machine learning and could be a sign of where the industry is headed. Two days before the HuggingFace funding announcement, open-source machine learning platform MetaSpore released a demo based on the HuggingFace Rapid deployment pre-training model. As deep learning technology makes innovative breakthroughs in computer vision, natural language processing, speech understanding, and other fields, more and more unstructured data are perceived, understood, and processed by machines. These advances are mainly due to the powerful learning ability of deep learning. Through pre-training of deep models on massive data, the models can capture the internal data patterns, thus helping many downstream tasks. With the industry and academia investing more and more energy in the research of pre-training technology, the distribution warehouses of pre-training models such as HuggingFace and Timm have emerged one after another. The open-source community release pre-training significant model dividends at an unprecedented speed. In recent years, the data form of machine modeling and understanding has gradually evolved from single-mode to multi-mode, and the semantic gap between different modes is being eliminated, making it possible to retrieve data across modes. Take CLIP, OpenAI’s open-source work, as an example, to pre-train the twin towers of images and texts on a dataset of 400 million pictures and texts and connect the semantics between pictures and texts. Many researchers in the academic world have been solving multimodal problems such as image generation and retrieval based on this technology. Although the frontier technology through the semantic gap between modal data, there is still a heavy and complicated model tuning, offline data processing, high performance online reasoning architecture design, heterogeneous computing, and online algorithm be born multiple processes and challenges, hindering the frontier multimodal retrieval technologies fall to the ground and pratt &whitney. DMetaSoul aims at the above technical pain points, abstracting and uniting many links such as model training optimization, online reasoning, and algorithm experiment, forming a set of solutions that can quickly apply offline pre-training model to online. This paper will introduce how to use the HuggingFace community pre-training model to conduct online reasoning and algorithm experiments based on MetaSpore technology ecology so that the benefits of the pre-training model can be fully released to the specific business or industry and small and medium-sized enterprises. And we will give the text search text and text search graph two multimodal retrieval demonstration examples for your reference. Multimodal semantic retrieval The sample architecture of multimodal retrieval is as follows: Our multimodal retrieval system supports both text search and text search application scenarios, including offline processing, model reasoning, online services, and other core modules: https://preview.redd.it/mdyyv1qmdz291.png?width=1834&format=png&auto=webp&s=e9e10710794c78c64cc05adb75db385aa53aba40 Offline processing, including offline data processing processes for different application scenarios of text search and text search, including model tuning, model export, data index database construction, data push, etc. Model inference. After the offline model training, we deployed our NLP and CV large models based on the MetaSpore Serving framework. MetaSpore Serving helps us conveniently perform online inference, elastic scheduling, load balancing, and resource scheduling in heterogeneous environments. Online services. Based on MetaSpore’s online algorithm application framework, MetaSpore has a complete set of reusable online search services, including Front-end retrieval UI, multimodal data preprocessing, vector recall and sorting algorithm, AB experimental framework, etc. MetaSpore also supports text search by text and image scene search by text and can be migrated to other application scenarios at a low cost. The HuggingFace open source community has provided several excellent baseline models for similar multimodal retrieval problems, which are often the starting point for actual optimization in the industry. MetaSpore also uses the pre-training model of the HuggingFace community in its online services of searching words by words and images by words. Searching words by words is based on the semantic similarity model of the question and answer field optimized by MetaSpore, and searching images by words is based on the community pre-training model. These community open source pre-training models are exported to the general ONNX format and loaded into MetaSpore Serving for online reasoning. The following sections will provide a detailed description of the model export and online retrieval algorithm services. The reasoning part of the model is standardized SAAS services with low coupling with the business. Interested readers can refer to my previous post: The design concept of MetaSpore, a new generation of the one-stop machine learning platform. 1.1 Offline Processing Offline processing mainly involves the export and loading of online models and index building and pushing of the document library. You can follow the step-by-step instructions below to complete the offline processing of text search and image search and see how the offline pre-training model achieves reasoning at MetaSpore. 1.1.1 Search text by text Traditional text retrieval systems are based on literal matching algorithms such as BM25. Due to users’ diverse query words, a semantic gap between query words and documents is often encountered. For example, users misspell “iPhone” as “Phone,” and search terms are incredibly long, such as “1 \~ 3 months old baby autumn small size bag pants”. Traditional text retrieval systems will use spelling correction, synonym expansion, search terms rewriting, and other means to alleviate the semantic gap but fundamentally fail to solve this problem. Only when the retrieval system fully understands users’ query terms and documents can it meet users’ retrieval demands at the semantic level. With the continuous progress of pre-training and representational learning technology, some commercial search engines continue to integrate semantic vector retrieval methods based on symbolic learning into the retrieval ecology. Semantic retrieval model This paper introduces a set of semantic vector retrieval applications. MetaSpore built a set of semantic retrieval systems based on encyclopedia question and answer data. MetaSpore adopted the Sentence-Bert model as the semantic vector representation model, which fine-tunes the twin tower BERT in supervised or unsupervised ways to make the model more suitable for retrieval tasks. The model structure is as follows: The query-Doc symmetric two-tower model is used in text search and question and answer retrieval. The vector representation of online Query and offline DOC share the same vector representation model, so it is necessary to ensure the consistency of the offline DOC library building model and online Query inference model. The case uses MetaSpore’s text representation model Sbert-Chinese-QMC-domain-V1, optimized in the open-source semantically similar data set. This model will express the question and answer data as a vector in offline database construction. The user query will be expressed as a vector by this model in online retrieval, ensuring that query-doc in the same semantic space, users’ semantic retrieval demands can be guaranteed by vector similarity metric calculation. Since the text presentation model does vector encoding for Query online, we need to export the model for use by the online service. Go to the q&A data library code directory and export the model concerning the documentation. In the script, Pytorch Tracing is used to export the model. The models are exported to the “./export “directory. The exported models are mainly ONNX models used for wired reasoning, Tokenizer, and related configuration files. The exported models are loaded into MetaSpore Serving by the online Serving system described below for model reasoning. Since the exported model will be copied to the cloud storage, you need to configure related variables in env.sh. \Build library based on text search \ The retrieval database is built on the million-level encyclopedia question and answer data set. According to the description document, you need to download the data and complete the database construction. The question and answer data will be coded as a vector by the offline model, and then the database construction data will be pushed to the service component. The whole process of database construction is described as follows: Preprocessing, converting the original data into a more general JSonline format for database construction; Build index, use the same model as online “sbert-Chinese-qmc-domain-v1” to index documents (one document object per line); Push inverted (vector) and forward (document field) data to each component server. The following is an example of the database data format. After offline database construction is completed, various data are pushed to corresponding service components, such as Milvus storing vector representation of documents and MongoDB storing summary information of documents. Online retrieval algorithm services will use these service components to obtain relevant data. 1.1.2 Search by text Text and images are easy for humans to relate semantically but difficult for machines. First of all, from the perspective of data form, the text is the discrete ID type of one-dimensional data based on words and words. At the same time, images are continuous two-dimensional or three-dimensional data. Secondly, the text is a subjective creation of human beings, and its expressive ability is vibrant, including various turning points, metaphors, and other expressions, while images are machine representations of the objective world. In short, bridging the semantic gap between text and image data is much more complex than searching text by text. The traditional text search image retrieval technology generally relies on the external text description data of the image or the nearest neighbor retrieval technology and carries out the retrieval through the image associated text, which in essence degrades the problem to text search. However, it will also face many issues, such as obtaining the associated text of pictures and whether the accuracy of text search by text is high enough. The depth model has gradually evolved from single-mode to multi-mode in recent years. Taking the open-source project of OpenAI, CLIP, as an example, train the model through the massive image and text data of the Internet and map the text and image data into the same semantic space, making it possible to implement the text and image search technology based on semantic vector. CLIP graphic model The text search pictures introduced in this paper are implemented based on semantic vector retrieval, and the CLIP pre-training model is used as the two-tower retrieval architecture. Because the CLIP model has trained the semantic alignment of the twin towers’ text and image side models on the massive graphic and text data, it is particularly suitable for the text search graph scene. Due to the different image and text data forms, the Query-Doc asymmetric twin towers model is used for text search image retrieval. The image-side model of the twin towers is used for offline database construction, and the text-side model is used for the online return. In the final online retrieval, the database data of the image side model will be searched after the text side model encodes Query, and the CLIP pre-training model guarantees the semantic correlation between images and texts. The model can draw the graphic pairs closer in vector space by pre-training on a large amount of visual data. Here we need to export the text-side model for online MetaSpore Serving inference. Since the retrieval scene is based on Chinese, the CLIP model supporting Chinese understanding is selected. The exported content includes the ONNX model used for online reasoning and Tokenizer, similar to the text search. MetaSpore Serving can load model reasoning through the exported content. Build library on Image search You need to download the Unsplash Lite library data and complete the construction according to the instructions. The whole process of database construction is described as follows: Preprocessing, specify the image directory, and then generate a more general JSOnline file for library construction; Build index, use OpenAI/Clip-Vit-BASE-Patch32 pre-training model to index the gallery, and output one document object for each line of index data; Push inverted (vector) and forward (document field) data to each component server. Like text search, after offline database construction, relevant data will be pushed to service components, called by online retrieval algorithm services to obtain relevant data. 1.2 Online Services The overall online service architecture diagram is as follows:  https://preview.redd.it/nz8zrbbpdz291.png?width=1280&format=png&auto=webp&s=28dae7e031621bc8819519667ed03d8d085d8ace Multi-mode search online service system supports application scenarios such as text search and text search. The whole online service consists of the following parts: Query preprocessing service: encapsulate preprocessing logic (including text/image, etc.) of pre-training model, and provide services through gRPC interface; Retrieval algorithm service: the whole algorithm processing link includes AB experiment tangent flow configuration, MetaSpore Serving call, vector recall, sorting, document summary, etc.; User entry service: provides a Web UI interface for users to debug and track down problems in the retrieval service. From a user request perspective, these services form invocation dependencies from back to front, so to build up a multimodal sample, you need to run each service from front to back first. Before doing this, remember to export the offline model, put it online and build the library first. This article will introduce the various parts of the online service system and make the whole service system step by step according to the following guidance. See the ReadME at the end of this article for more details. 1.2.1 Query preprocessing service Deep learning models tend to be based on tensors, but NLP/CV models often have a preprocessing part that translates raw text and images into tensors that deep learning models can accept. For example, NLP class models often have a pre-tokenizer to transform text data of string type into discrete tensor data. CV class models also have similar processing logic to complete the cropping, scaling, transformation, and other processing of input images through preprocessing. On the one hand, considering that this part of preprocessing logic is decoupled from tensor reasoning of the depth model, on the other hand, the reason of the depth model has an independent technical system based on ONNX, so MetaSpore disassembled this part of preprocessing logic. NLP pretreatment Tokenizer has been integrated into the Query pretreatment service. MetaSpore dismantlement with a relatively general convention. Users only need to provide preprocessing logic files to realize the loading and prediction interface and export the necessary data and configuration files loaded into the preprocessing service. Subsequent CV preprocessing logic will also be integrated in this manner. The preprocessing service currently provides the gRPC interface invocation externally and is dependent on the Query preprocessing (QP) module in the retrieval algorithm service. After the user request reaches the retrieval algorithm service, it will be forwarded to the service to complete the data preprocessing and continue the subsequent processing. The ReadMe provides details on how the preprocessing service is started, how the preprocessing model exported offline to cloud storage enters the service, and how to debug the service. To further improve the efficiency and stability of model reasoning, MetaSpore Serving implements a Python preprocessing submodule. So MetaSpore can provide gRPC services through user-specified preprocessor.py, complete Tokenizer or CV-related preprocessing in NLP, and translate requests into a Tensor that deep models can handle. Finally, the model inference is carried out by MetaSpore, Serving subsequent sub-modules. Presented here on the lot code: https://github.com/meta-soul/MetaSpore/compare/add\python\preprocessor 1.2.2 Retrieval algorithm services Retrieval algorithm service is the core of the whole online service system, which is responsible for the triage of experiments, the assembly of algorithm chains such as preprocessing, recall, sorting, and the invocation of dependent component services. The whole retrieval algorithm service is developed based on the Java Spring framework and supports multi-mode retrieval scenarios of text search and text search graph. Due to good internal abstraction and modular design, it has high flexibility and can be migrated to similar application scenarios at a low cost. Here’s a quick guide to configuring the environment to set up the retrieval algorithm service. See ReadME for more details: Install dependent components. Use Maven to install the online-Serving component Search for service configurations. Copy the template configuration file and replace the MongoDB, Milvus, and other configurations based on the development/production environment. Install and configure Consul. Consul allows you to synchronize the search service configuration in real-time, including cutting the flow of experiments, recall parameters, and sorting parameters. The project’s configuration file shows the current configuration parameters of text search and text search. The parameter modelName in the stage of pretreatment and recall is the corresponding model exported in offline processing. Start the service. Once the above configuration is complete, the retrieval service can be started from the entry script. Once the service is started, you can test it! For example, for a user with userId=10 who wants to query “How to renew ID card,” access the text search service. 1.2.3 User Entry Service Considering that the retrieval algorithm service is in the form of the API interface, it is difficult to locate and trace the problem, especially for the text search image scene can intuitively display the retrieval results to facilitate the iterative optimization of the retrieval algorithm. This paper provides a lightweight Web UI interface for text search and image search, a search input box, and results in a display page for users. Developed by Flask, the service can be easily integrated with other retrieval applications. The service calls the retrieval algorithm service and displays the returned results on the page. It’s also easy to install and start the service. Once you’re done, go to http://127.0.0.1:8090 to see if the search UI service is working correctly. See the ReadME at the end of this article for details. Multimodal system demonstration The multimodal retrieval service can be started when offline processing and online service environment configuration have been completed following the above instructions. Examples of textual searches are shown below. Enter the entry of the text search map application, enter “cat” first, and you can see that the first three digits of the returned result are cats: https://preview.redd.it/d7syq47rdz291.png?width=1280&format=png&auto=webp&s=b43df9abd380b7d9a52e3045dd787f4feeb69635 If you add a color constraint to “cat” to retrieve “black cat,” you can see that it does return a black cat:  https://preview.redd.it/aa7pxx8tdz291.png?width=1280&format=png&auto=webp&s=e3727c29d1bde6eea2e1cccf6c46d3cae3f4750e Further, strengthen the constraint on the search term, change it to “black cat on the bed,” and return results containing pictures of a black cat climbing on the bed:  https://preview.redd.it/2mw4qpjudz291.png?width=1280&format=png&auto=webp&s=1cf1db667892b9b3a40451993680fbd6980b5520 The cat can still be found through the text search system after the color and scene modification in the above example. Conclusion The cutting-edge pre-training technology can bridge the semantic gap between different modes, and the HuggingFace community can greatly reduce the cost for developers to use the pre-training model. Combined with the technological ecology of MetaSpore online reasoning and online microservices provided by DMetaSpore, the pre-training model is no longer mere offline dabbling. Instead, it can truly achieve end-to-end implementation from cutting-edge technology to industrial scenarios, fully releasing the dividends of the pre-training large model. In the future, DMetaSoul will continue to improve and optimize the MetaSpore technology ecosystem: More automated and wider access to HuggingFace community ecology. MetaSpore will soon release a common model rollout mechanism to make HuggingFace ecologically accessible and will later integrate preprocessing services into online services. Multi-mode retrieval offline algorithm optimization. For multimodal retrieval scenarios, MetaSpore will continuously iteratively optimize offline algorithm components, including text recall/sort model, graphic recall/sort model, etc., to improve the accuracy and efficiency of the retrieval algorithm. For related code and reference documentation in this article, please visit: https://github.com/meta-soul/MetaSpore/tree/main/demo/multimodal/online Some images source: https://github.com/openai/CLIP/raw/main/CLIP.png https://www.sbert.net/examples/training/sts/README.html

#reddit #mmml #deploy

I made a website to generate craft ideas from existing materials

0

tiklook•This week

I made a website to generate craft ideas from existing materials

Idea: When browsing crafts groups, I noticed that members searched for craft ideas, either randomly, using available materials, or recycling existing things: First attempt: This attempt was a failure. After studying the idea well, I started trying to implement it with the help of Gpt. Because I am a beginner in programming, I only have a little experience in. html and css. I used gpt for php and js. The goal of the site was for the user to enter the materials he owns and the degree of difficulty of implementation to get a title for the idea. And how to implement it. This was all writing generated through the open ai api. After a few struggles, it became clear to me that writing could not describe the idea as it is. I changed the idea. Ideas are generated using images. The results were amazing. Here I faced a budget problem, as the subscriptions to the API for images were very high in my opinion. Because I plan to make the tool free.So I left the project. Second attempt. A year later while I was browsing I discovered stability.ai. From stablediffision, I found their prices to be reasonable, so I returned to the project and completed it. Website link: Craftivi

#reddit #made #website

[D] What is your honest experience with reinforcement learning?

0

1

Starks-Technology•This week

[D] What is your honest experience with reinforcement learning?

In my personal experience, SOTA RL algorithms simply don't work. I've tried working with reinforcement learning for over 5 years. I remember when Alpha Go defeated the world famous Go player, Lee Sedol, and everybody thought RL would take the ML community by storm. Yet, outside of toy problems, I've personally never found a practical use-case of RL. What is your experience with it? Aside from Ad recommendation systems and RLHF, are there legitimate use-cases of RL? Or, was it all hype? Edit: I know a lot about AI. I built NexusTrade, an AI-Powered automated investing tool that lets non-technical users create, update, and deploy their trading strategies. I’m not an idiot nor a noob; RL is just ridiculously hard. Edit 2: Since my comments are being downvoted, here is a link to my article that better describes my position. It's not that I don't understand RL. I released my open-source code and wrote a paper on it. It's the fact that it's EXTREMELY difficult to understand. Other deep learning algorithms like CNNs (including ResNets), RNNs (including GRUs and LSTMs), Transformers, and GANs are not hard to understand. These algorithms work and have practical use-cases outside of the lab. Traditional SOTA RL algorithms like PPO, DDPG, and TD3 are just very hard. You need to do a bunch of research to even implement a toy problem. In contrast, the decision transformer is something anybody can implement, and it seems to match or surpass the SOTA. You don't need two networks battling each other. You don't have to go through hell to debug your network. It just naturally learns the best set of actions in an auto-regressive manner. I also didn't mean to come off as arrogant or imply that RL is not worth learning. I just haven't seen any real-world, practical use-cases of it. I simply wanted to start a discussion, not claim that I know everything. Edit 3: There's a shockingly number of people calling me an idiot for not fully understanding RL. You guys are wayyy too comfortable calling people you disagree with names. News-flash, not everybody has a PhD in ML. My undergraduate degree is in biology. I self-taught myself the high-level maths to understand ML. I'm very passionate about the field; I just have VERY disappointing experiences with RL. Funny enough, there are very few people refuting my actual points. To summarize: Lack of real-world applications Extremely complex and inaccessible to 99% of the population Much harder than traditional DL algorithms like CNNs, RNNs, and GANs Sample inefficiency and instability Difficult to debug Better alternatives, such as the Decision Transformer Are these not legitimate criticisms? Is the purpose of this sub not to have discussions related to Machine Learning? To the few commenters that aren't calling me an idiot...thank you! Remember, it costs you nothing to be nice! Edit 4: Lots of people seem to agree that RL is over-hyped. Unfortunately those comments are downvoted. To clear up some things: We've invested HEAVILY into reinforcement learning. All we got from this investment is a robot that can be super-human at (some) video games. AlphaFold did not use any reinforcement learning. SpaceX doesn't either. I concede that it can be useful for robotics, but still argue that it's use-cases outside the lab are extremely limited. If you're stumbling on this thread and curious about an RL alternative, check out the Decision Transformer. It can be used in any situation that a traditional RL algorithm can be used. Final Edit: To those who contributed more recently, thank you for the thoughtful discussion! From what I learned, model-based models like Dreamer and IRIS MIGHT have a future. But everybody who has actually used model-free models like DDPG unanimously agree that they suck and don’t work.

#reddit #your #honest

0

1

Starks-Technology•This week

[D] What is your honest experience with reinforcement learning?

In my personal experience, SOTA RL algorithms simply don't work. I've tried working with reinforcement learning for over 5 years. I remember when Alpha Go defeated the world famous Go player, Lee Sedol, and everybody thought RL would take the ML community by storm. Yet, outside of toy problems, I've personally never found a practical use-case of RL. What is your experience with it? Aside from Ad recommendation systems and RLHF, are there legitimate use-cases of RL? Or, was it all hype? Edit: I know a lot about AI. I built NexusTrade, an AI-Powered automated investing tool that lets non-technical users create, update, and deploy their trading strategies. I’m not an idiot nor a noob; RL is just ridiculously hard. Edit 2: Since my comments are being downvoted, here is a link to my article that better describes my position. It's not that I don't understand RL. I released my open-source code and wrote a paper on it. It's the fact that it's EXTREMELY difficult to understand. Other deep learning algorithms like CNNs (including ResNets), RNNs (including GRUs and LSTMs), Transformers, and GANs are not hard to understand. These algorithms work and have practical use-cases outside of the lab. Traditional SOTA RL algorithms like PPO, DDPG, and TD3 are just very hard. You need to do a bunch of research to even implement a toy problem. In contrast, the decision transformer is something anybody can implement, and it seems to match or surpass the SOTA. You don't need two networks battling each other. You don't have to go through hell to debug your network. It just naturally learns the best set of actions in an auto-regressive manner. I also didn't mean to come off as arrogant or imply that RL is not worth learning. I just haven't seen any real-world, practical use-cases of it. I simply wanted to start a discussion, not claim that I know everything. Edit 3: There's a shockingly number of people calling me an idiot for not fully understanding RL. You guys are wayyy too comfortable calling people you disagree with names. News-flash, not everybody has a PhD in ML. My undergraduate degree is in biology. I self-taught myself the high-level maths to understand ML. I'm very passionate about the field; I just have VERY disappointing experiences with RL. Funny enough, there are very few people refuting my actual points. To summarize: Lack of real-world applications Extremely complex and inaccessible to 99% of the population Much harder than traditional DL algorithms like CNNs, RNNs, and GANs Sample inefficiency and instability Difficult to debug Better alternatives, such as the Decision Transformer Are these not legitimate criticisms? Is the purpose of this sub not to have discussions related to Machine Learning? To the few commenters that aren't calling me an idiot...thank you! Remember, it costs you nothing to be nice! Edit 4: Lots of people seem to agree that RL is over-hyped. Unfortunately those comments are downvoted. To clear up some things: We've invested HEAVILY into reinforcement learning. All we got from this investment is a robot that can be super-human at (some) video games. AlphaFold did not use any reinforcement learning. SpaceX doesn't either. I concede that it can be useful for robotics, but still argue that it's use-cases outside the lab are extremely limited. If you're stumbling on this thread and curious about an RL alternative, check out the Decision Transformer. It can be used in any situation that a traditional RL algorithm can be used. Final Edit: To those who contributed more recently, thank you for the thoughtful discussion! From what I learned, model-based models like Dreamer and IRIS MIGHT have a future. But everybody who has actually used model-free models like DDPG unanimously agree that they suck and don’t work.

#reddit #your #honest

[P]MMML | Deploy HuggingFace training model rapidly based on MetaSpore

0

1

qazmkopp•This week

[P]MMML | Deploy HuggingFace training model rapidly based on MetaSpore

A few days ago, HuggingFace announced a $100 million Series C funding round, which was big news in open source machine learning and could be a sign of where the industry is headed. Two days before the HuggingFace funding announcement, open-source machine learning platform MetaSpore released a demo based on the HuggingFace Rapid deployment pre-training model. As deep learning technology makes innovative breakthroughs in computer vision, natural language processing, speech understanding, and other fields, more and more unstructured data are perceived, understood, and processed by machines. These advances are mainly due to the powerful learning ability of deep learning. Through pre-training of deep models on massive data, the models can capture the internal data patterns, thus helping many downstream tasks. With the industry and academia investing more and more energy in the research of pre-training technology, the distribution warehouses of pre-training models such as HuggingFace and Timm have emerged one after another. The open-source community release pre-training significant model dividends at an unprecedented speed. In recent years, the data form of machine modeling and understanding has gradually evolved from single-mode to multi-mode, and the semantic gap between different modes is being eliminated, making it possible to retrieve data across modes. Take CLIP, OpenAI’s open-source work, as an example, to pre-train the twin towers of images and texts on a dataset of 400 million pictures and texts and connect the semantics between pictures and texts. Many researchers in the academic world have been solving multimodal problems such as image generation and retrieval based on this technology. Although the frontier technology through the semantic gap between modal data, there is still a heavy and complicated model tuning, offline data processing, high performance online reasoning architecture design, heterogeneous computing, and online algorithm be born multiple processes and challenges, hindering the frontier multimodal retrieval technologies fall to the ground and pratt &whitney. DMetaSoul aims at the above technical pain points, abstracting and uniting many links such as model training optimization, online reasoning, and algorithm experiment, forming a set of solutions that can quickly apply offline pre-training model to online. This paper will introduce how to use the HuggingFace community pre-training model to conduct online reasoning and algorithm experiments based on MetaSpore technology ecology so that the benefits of the pre-training model can be fully released to the specific business or industry and small and medium-sized enterprises. And we will give the text search text and text search graph two multimodal retrieval demonstration examples for your reference. Multimodal semantic retrieval The sample architecture of multimodal retrieval is as follows: Our multimodal retrieval system supports both text search and text search application scenarios, including offline processing, model reasoning, online services, and other core modules:  https://preview.redd.it/w4v4c7vcez291.png?width=1834&format=png&auto=webp&s=0687efb1fddb26e8e30cb844d398ec712b947f31 Offline processing, including offline data processing processes for different application scenarios of text search and text search, including model tuning, model export, data index database construction, data push, etc. Model inference. After the offline model training, we deployed our NLP and CV large models based on the MetaSpore Serving framework. MetaSpore Serving helps us conveniently perform online inference, elastic scheduling, load balancing, and resource scheduling in heterogeneous environments. Online services. Based on MetaSpore’s online algorithm application framework, MetaSpore has a complete set of reusable online search services, including Front-end retrieval UI, multimodal data preprocessing, vector recall and sorting algorithm, AB experimental framework, etc. MetaSpore also supports text search by text and image scene search by text and can be migrated to other application scenarios at a low cost. The HuggingFace open source community has provided several excellent baseline models for similar multimodal retrieval problems, which are often the starting point for actual optimization in the industry. MetaSpore also uses the pre-training model of the HuggingFace community in its online services of searching words by words and images by words. Searching words by words is based on the semantic similarity model of the question and answer field optimized by MetaSpore, and searching images by words is based on the community pre-training model. These community open source pre-training models are exported to the general ONNX format and loaded into MetaSpore Serving for online reasoning. The following sections will provide a detailed description of the model export and online retrieval algorithm services. The reasoning part of the model is standardized SAAS services with low coupling with the business. Interested readers can refer to my previous post: The design concept of MetaSpore, a new generation of the one-stop machine learning platform. 1.1 Offline Processing Offline processing mainly involves the export and loading of online models and index building and pushing of the document library. You can follow the step-by-step instructions below to complete the offline processing of text search and image search and see how the offline pre-training model achieves reasoning at MetaSpore. 1.1.1 Search text by text Traditional text retrieval systems are based on literal matching algorithms such as BM25. Due to users’ diverse query words, a semantic gap between query words and documents is often encountered. For example, users misspell “iPhone” as “Phone,” and search terms are incredibly long, such as “1 \~ 3 months old baby autumn small size bag pants”. Traditional text retrieval systems will use spelling correction, synonym expansion, search terms rewriting, and other means to alleviate the semantic gap but fundamentally fail to solve this problem. Only when the retrieval system fully understands users’ query terms and documents can it meet users’ retrieval demands at the semantic level. With the continuous progress of pre-training and representational learning technology, some commercial search engines continue to integrate semantic vector retrieval methods based on symbolic learning into the retrieval ecology. Semantic retrieval model This paper introduces a set of semantic vector retrieval applications. MetaSpore built a set of semantic retrieval systems based on encyclopedia question and answer data. MetaSpore adopted the Sentence-Bert model as the semantic vector representation model, which fine-tunes the twin tower BERT in supervised or unsupervised ways to make the model more suitable for retrieval tasks. The model structure is as follows: The query-Doc symmetric two-tower model is used in text search and question and answer retrieval. The vector representation of online Query and offline DOC share the same vector representation model, so it is necessary to ensure the consistency of the offline DOC library building model and online Query inference model. The case uses MetaSpore’s text representation model Sbert-Chinese-QMC-domain-V1, optimized in the open-source semantically similar data set. This model will express the question and answer data as a vector in offline database construction. The user query will be expressed as a vector by this model in online retrieval, ensuring that query-doc in the same semantic space, users’ semantic retrieval demands can be guaranteed by vector similarity metric calculation. Since the text presentation model does vector encoding for Query online, we need to export the model for use by the online service. Go to the q&A data library code directory and export the model concerning the documentation. In the script, Pytorch Tracing is used to export the model. The models are exported to the “./export “directory. The exported models are mainly ONNX models used for wired reasoning, Tokenizer, and related configuration files. The exported models are loaded into MetaSpore Serving by the online Serving system described below for model reasoning. Since the exported model will be copied to the cloud storage, you need to configure related variables in env.sh. \Build library based on text search \ The retrieval database is built on the million-level encyclopedia question and answer data set. According to the description document, you need to download the data and complete the database construction. The question and answer data will be coded as a vector by the offline model, and then the database construction data will be pushed to the service component. The whole process of database construction is described as follows: Preprocessing, converting the original data into a more general JSonline format for database construction; Build index, use the same model as online “sbert-Chinese-qmc-domain-v1” to index documents (one document object per line); Push inverted (vector) and forward (document field) data to each component server. The following is an example of the database data format. After offline database construction is completed, various data are pushed to corresponding service components, such as Milvus storing vector representation of documents and MongoDB storing summary information of documents. Online retrieval algorithm services will use these service components to obtain relevant data. 1.1.2 Search by text Text and images are easy for humans to relate semantically but difficult for machines. First of all, from the perspective of data form, the text is the discrete ID type of one-dimensional data based on words and words. At the same time, images are continuous two-dimensional or three-dimensional data. Secondly, the text is a subjective creation of human beings, and its expressive ability is vibrant, including various turning points, metaphors, and other expressions, while images are machine representations of the objective world. In short, bridging the semantic gap between text and image data is much more complex than searching text by text. The traditional text search image retrieval technology generally relies on the external text description data of the image or the nearest neighbor retrieval technology and carries out the retrieval through the image associated text, which in essence degrades the problem to text search. However, it will also face many issues, such as obtaining the associated text of pictures and whether the accuracy of text search by text is high enough. The depth model has gradually evolved from single-mode to multi-mode in recent years. Taking the open-source project of OpenAI, CLIP, as an example, train the model through the massive image and text data of the Internet and map the text and image data into the same semantic space, making it possible to implement the text and image search technology based on semantic vector. CLIP graphic model The text search pictures introduced in this paper are implemented based on semantic vector retrieval, and the CLIP pre-training model is used as the two-tower retrieval architecture. Because the CLIP model has trained the semantic alignment of the twin towers’ text and image side models on the massive graphic and text data, it is particularly suitable for the text search graph scene. Due to the different image and text data forms, the Query-Doc asymmetric twin towers model is used for text search image retrieval. The image-side model of the twin towers is used for offline database construction, and the text-side model is used for the online return. In the final online retrieval, the database data of the image side model will be searched after the text side model encodes Query, and the CLIP pre-training model guarantees the semantic correlation between images and texts. The model can draw the graphic pairs closer in vector space by pre-training on a large amount of visual data. Here we need to export the text-side model for online MetaSpore Serving inference. Since the retrieval scene is based on Chinese, the CLIP model supporting Chinese understanding is selected. The exported content includes the ONNX model used for online reasoning and Tokenizer, similar to the text search. MetaSpore Serving can load model reasoning through the exported content. Build library on Image search You need to download the Unsplash Lite library data and complete the construction according to the instructions. The whole process of database construction is described as follows: Preprocessing, specify the image directory, and then generate a more general JSOnline file for library construction; Build index, use OpenAI/Clip-Vit-BASE-Patch32 pre-training model to index the gallery, and output one document object for each line of index data; Push inverted (vector) and forward (document field) data to each component server. Like text search, after offline database construction, relevant data will be pushed to service components, called by online retrieval algorithm services to obtain relevant data. 1.2 Online Services The overall online service architecture diagram is as follows: https://preview.redd.it/jfsl8hdfez291.png?width=1280&format=png&auto=webp&s=a858e2304a0c93e78ba5429612ca08cbee69b35a Multi-mode search online service system supports application scenarios such as text search and text search. The whole online service consists of the following parts: Query preprocessing service: encapsulate preprocessing logic (including text/image, etc.) of pre-training model, and provide services through gRPC interface; Retrieval algorithm service: the whole algorithm processing link includes AB experiment tangent flow configuration, MetaSpore Serving call, vector recall, sorting, document summary, etc.; User entry service: provides a Web UI interface for users to debug and track down problems in the retrieval service. From a user request perspective, these services form invocation dependencies from back to front, so to build up a multimodal sample, you need to run each service from front to back first. Before doing this, remember to export the offline model, put it online and build the library first. This article will introduce the various parts of the online service system and make the whole service system step by step according to the following guidance. See the ReadME at the end of this article for more details. 1.2.1 Query preprocessing service Deep learning models tend to be based on tensors, but NLP/CV models often have a preprocessing part that translates raw text and images into tensors that deep learning models can accept. For example, NLP class models often have a pre-tokenizer to transform text data of string type into discrete tensor data. CV class models also have similar processing logic to complete the cropping, scaling, transformation, and other processing of input images through preprocessing. On the one hand, considering that this part of preprocessing logic is decoupled from tensor reasoning of the depth model, on the other hand, the reason of the depth model has an independent technical system based on ONNX, so MetaSpore disassembled this part of preprocessing logic. NLP pretreatment Tokenizer has been integrated into the Query pretreatment service. MetaSpore dismantlement with a relatively general convention. Users only need to provide preprocessing logic files to realize the loading and prediction interface and export the necessary data and configuration files loaded into the preprocessing service. Subsequent CV preprocessing logic will also be integrated in this manner. The preprocessing service currently provides the gRPC interface invocation externally and is dependent on the Query preprocessing (QP) module in the retrieval algorithm service. After the user request reaches the retrieval algorithm service, it will be forwarded to the service to complete the data preprocessing and continue the subsequent processing. The ReadMe provides details on how the preprocessing service is started, how the preprocessing model exported offline to cloud storage enters the service, and how to debug the service. To further improve the efficiency and stability of model reasoning, MetaSpore Serving implements a Python preprocessing submodule. So MetaSpore can provide gRPC services through user-specified preprocessor.py, complete Tokenizer or CV-related preprocessing in NLP, and translate requests into a Tensor that deep models can handle. Finally, the model inference is carried out by MetaSpore, Serving subsequent sub-modules. Presented here on the lot code: https://github.com/meta-soul/MetaSpore/compare/add\python\preprocessor 1.2.2 Retrieval algorithm services Retrieval algorithm service is the core of the whole online service system, which is responsible for the triage of experiments, the assembly of algorithm chains such as preprocessing, recall, sorting, and the invocation of dependent component services. The whole retrieval algorithm service is developed based on the Java Spring framework and supports multi-mode retrieval scenarios of text search and text search graph. Due to good internal abstraction and modular design, it has high flexibility and can be migrated to similar application scenarios at a low cost. Here’s a quick guide to configuring the environment to set up the retrieval algorithm service. See ReadME for more details: Install dependent components. Use Maven to install the online-Serving component Search for service configurations. Copy the template configuration file and replace the MongoDB, Milvus, and other configurations based on the development/production environment. Install and configure Consul. Consul allows you to synchronize the search service configuration in real-time, including cutting the flow of experiments, recall parameters, and sorting parameters. The project’s configuration file shows the current configuration parameters of text search and text search. The parameter modelName in the stage of pretreatment and recall is the corresponding model exported in offline processing. Start the service. Once the above configuration is complete, the retrieval service can be started from the entry script. Once the service is started, you can test it! For example, for a user with userId=10 who wants to query “How to renew ID card,” access the text search service. 1.2.3 User Entry Service Considering that the retrieval algorithm service is in the form of the API interface, it is difficult to locate and trace the problem, especially for the text search image scene can intuitively display the retrieval results to facilitate the iterative optimization of the retrieval algorithm. This paper provides a lightweight Web UI interface for text search and image search, a search input box, and results in a display page for users. Developed by Flask, the service can be easily integrated with other retrieval applications. The service calls the retrieval algorithm service and displays the returned results on the page. It’s also easy to install and start the service. Once you’re done, go to http://127.0.0.1:8090 to see if the search UI service is working correctly. See the ReadME at the end of this article for details. Multimodal system demonstration The multimodal retrieval service can be started when offline processing and online service environment configuration have been completed following the above instructions. Examples of textual searches are shown below. Enter the entry of the text search map application, enter “cat” first, and you can see that the first three digits of the returned result are cats: https://preview.redd.it/0n5nuyvhez291.png?width=1280&format=png&auto=webp&s=1e9c054f541d53381674b8d6001b4bf524506bd2 If you add a color constraint to “cat” to retrieve “black cat,” you can see that it does return a black cat: https://preview.redd.it/rzc0qjyjez291.png?width=1280&format=png&auto=webp&s=d5bcc503ef0fb3360c7740e60e295cf372dcad47 Further, strengthen the constraint on the search term, change it to “black cat on the bed,” and return results containing pictures of a black cat climbing on the bed:  https://preview.redd.it/c4b2q8olez291.png?width=1280&format=png&auto=webp&s=4f3817b0b9f07e1e68d1d4a8281702ba3834a00a The cat can still be found through the text search system after the color and scene modification in the above example. Conclusion The cutting-edge pre-training technology can bridge the semantic gap between different modes, and the HuggingFace community can greatly reduce the cost for developers to use the pre-training model. Combined with the technological ecology of MetaSpore online reasoning and online microservices provided by DMetaSpore, the pre-training model is no longer mere offline dabbling. Instead, it can truly achieve end-to-end implementation from cutting-edge technology to industrial scenarios, fully releasing the dividends of the pre-training large model. In the future, DMetaSoul will continue to improve and optimize the MetaSpore technology ecosystem: More automated and wider access to HuggingFace community ecology. MetaSpore will soon release a common model rollout mechanism to make HuggingFace ecologically accessible and will later integrate preprocessing services into online services. Multi-mode retrieval offline algorithm optimization. For multimodal retrieval scenarios, MetaSpore will continuously iteratively optimize offline algorithm components, including text recall/sort model, graphic recall/sort model, etc., to improve the accuracy and efficiency of the retrieval algorithm. For related code and reference documentation in this article, please visit: https://github.com/meta-soul/MetaSpore/tree/main/demo/multimodal/online Some images source: https://github.com/openai/CLIP/raw/main/CLIP.png https://www.sbert.net/examples/training/sts/README.html

#reddit #[p]mmml #deploy

0

1

qazmkopp•This week

[P]MMML | Deploy HuggingFace training model rapidly based on MetaSpore

A few days ago, HuggingFace announced a $100 million Series C funding round, which was big news in open source machine learning and could be a sign of where the industry is headed. Two days before the HuggingFace funding announcement, open-source machine learning platform MetaSpore released a demo based on the HuggingFace Rapid deployment pre-training model. As deep learning technology makes innovative breakthroughs in computer vision, natural language processing, speech understanding, and other fields, more and more unstructured data are perceived, understood, and processed by machines. These advances are mainly due to the powerful learning ability of deep learning. Through pre-training of deep models on massive data, the models can capture the internal data patterns, thus helping many downstream tasks. With the industry and academia investing more and more energy in the research of pre-training technology, the distribution warehouses of pre-training models such as HuggingFace and Timm have emerged one after another. The open-source community release pre-training significant model dividends at an unprecedented speed. In recent years, the data form of machine modeling and understanding has gradually evolved from single-mode to multi-mode, and the semantic gap between different modes is being eliminated, making it possible to retrieve data across modes. Take CLIP, OpenAI’s open-source work, as an example, to pre-train the twin towers of images and texts on a dataset of 400 million pictures and texts and connect the semantics between pictures and texts. Many researchers in the academic world have been solving multimodal problems such as image generation and retrieval based on this technology. Although the frontier technology through the semantic gap between modal data, there is still a heavy and complicated model tuning, offline data processing, high performance online reasoning architecture design, heterogeneous computing, and online algorithm be born multiple processes and challenges, hindering the frontier multimodal retrieval technologies fall to the ground and pratt &whitney. DMetaSoul aims at the above technical pain points, abstracting and uniting many links such as model training optimization, online reasoning, and algorithm experiment, forming a set of solutions that can quickly apply offline pre-training model to online. This paper will introduce how to use the HuggingFace community pre-training model to conduct online reasoning and algorithm experiments based on MetaSpore technology ecology so that the benefits of the pre-training model can be fully released to the specific business or industry and small and medium-sized enterprises. And we will give the text search text and text search graph two multimodal retrieval demonstration examples for your reference. Multimodal semantic retrieval The sample architecture of multimodal retrieval is as follows: Our multimodal retrieval system supports both text search and text search application scenarios, including offline processing, model reasoning, online services, and other core modules:  https://preview.redd.it/w4v4c7vcez291.png?width=1834&format=png&auto=webp&s=0687efb1fddb26e8e30cb844d398ec712b947f31 Offline processing, including offline data processing processes for different application scenarios of text search and text search, including model tuning, model export, data index database construction, data push, etc. Model inference. After the offline model training, we deployed our NLP and CV large models based on the MetaSpore Serving framework. MetaSpore Serving helps us conveniently perform online inference, elastic scheduling, load balancing, and resource scheduling in heterogeneous environments. Online services. Based on MetaSpore’s online algorithm application framework, MetaSpore has a complete set of reusable online search services, including Front-end retrieval UI, multimodal data preprocessing, vector recall and sorting algorithm, AB experimental framework, etc. MetaSpore also supports text search by text and image scene search by text and can be migrated to other application scenarios at a low cost. The HuggingFace open source community has provided several excellent baseline models for similar multimodal retrieval problems, which are often the starting point for actual optimization in the industry. MetaSpore also uses the pre-training model of the HuggingFace community in its online services of searching words by words and images by words. Searching words by words is based on the semantic similarity model of the question and answer field optimized by MetaSpore, and searching images by words is based on the community pre-training model. These community open source pre-training models are exported to the general ONNX format and loaded into MetaSpore Serving for online reasoning. The following sections will provide a detailed description of the model export and online retrieval algorithm services. The reasoning part of the model is standardized SAAS services with low coupling with the business. Interested readers can refer to my previous post: The design concept of MetaSpore, a new generation of the one-stop machine learning platform. 1.1 Offline Processing Offline processing mainly involves the export and loading of online models and index building and pushing of the document library. You can follow the step-by-step instructions below to complete the offline processing of text search and image search and see how the offline pre-training model achieves reasoning at MetaSpore. 1.1.1 Search text by text Traditional text retrieval systems are based on literal matching algorithms such as BM25. Due to users’ diverse query words, a semantic gap between query words and documents is often encountered. For example, users misspell “iPhone” as “Phone,” and search terms are incredibly long, such as “1 \~ 3 months old baby autumn small size bag pants”. Traditional text retrieval systems will use spelling correction, synonym expansion, search terms rewriting, and other means to alleviate the semantic gap but fundamentally fail to solve this problem. Only when the retrieval system fully understands users’ query terms and documents can it meet users’ retrieval demands at the semantic level. With the continuous progress of pre-training and representational learning technology, some commercial search engines continue to integrate semantic vector retrieval methods based on symbolic learning into the retrieval ecology. Semantic retrieval model This paper introduces a set of semantic vector retrieval applications. MetaSpore built a set of semantic retrieval systems based on encyclopedia question and answer data. MetaSpore adopted the Sentence-Bert model as the semantic vector representation model, which fine-tunes the twin tower BERT in supervised or unsupervised ways to make the model more suitable for retrieval tasks. The model structure is as follows: The query-Doc symmetric two-tower model is used in text search and question and answer retrieval. The vector representation of online Query and offline DOC share the same vector representation model, so it is necessary to ensure the consistency of the offline DOC library building model and online Query inference model. The case uses MetaSpore’s text representation model Sbert-Chinese-QMC-domain-V1, optimized in the open-source semantically similar data set. This model will express the question and answer data as a vector in offline database construction. The user query will be expressed as a vector by this model in online retrieval, ensuring that query-doc in the same semantic space, users’ semantic retrieval demands can be guaranteed by vector similarity metric calculation. Since the text presentation model does vector encoding for Query online, we need to export the model for use by the online service. Go to the q&A data library code directory and export the model concerning the documentation. In the script, Pytorch Tracing is used to export the model. The models are exported to the “./export “directory. The exported models are mainly ONNX models used for wired reasoning, Tokenizer, and related configuration files. The exported models are loaded into MetaSpore Serving by the online Serving system described below for model reasoning. Since the exported model will be copied to the cloud storage, you need to configure related variables in env.sh. \Build library based on text search \ The retrieval database is built on the million-level encyclopedia question and answer data set. According to the description document, you need to download the data and complete the database construction. The question and answer data will be coded as a vector by the offline model, and then the database construction data will be pushed to the service component. The whole process of database construction is described as follows: Preprocessing, converting the original data into a more general JSonline format for database construction; Build index, use the same model as online “sbert-Chinese-qmc-domain-v1” to index documents (one document object per line); Push inverted (vector) and forward (document field) data to each component server. The following is an example of the database data format. After offline database construction is completed, various data are pushed to corresponding service components, such as Milvus storing vector representation of documents and MongoDB storing summary information of documents. Online retrieval algorithm services will use these service components to obtain relevant data. 1.1.2 Search by text Text and images are easy for humans to relate semantically but difficult for machines. First of all, from the perspective of data form, the text is the discrete ID type of one-dimensional data based on words and words. At the same time, images are continuous two-dimensional or three-dimensional data. Secondly, the text is a subjective creation of human beings, and its expressive ability is vibrant, including various turning points, metaphors, and other expressions, while images are machine representations of the objective world. In short, bridging the semantic gap between text and image data is much more complex than searching text by text. The traditional text search image retrieval technology generally relies on the external text description data of the image or the nearest neighbor retrieval technology and carries out the retrieval through the image associated text, which in essence degrades the problem to text search. However, it will also face many issues, such as obtaining the associated text of pictures and whether the accuracy of text search by text is high enough. The depth model has gradually evolved from single-mode to multi-mode in recent years. Taking the open-source project of OpenAI, CLIP, as an example, train the model through the massive image and text data of the Internet and map the text and image data into the same semantic space, making it possible to implement the text and image search technology based on semantic vector. CLIP graphic model The text search pictures introduced in this paper are implemented based on semantic vector retrieval, and the CLIP pre-training model is used as the two-tower retrieval architecture. Because the CLIP model has trained the semantic alignment of the twin towers’ text and image side models on the massive graphic and text data, it is particularly suitable for the text search graph scene. Due to the different image and text data forms, the Query-Doc asymmetric twin towers model is used for text search image retrieval. The image-side model of the twin towers is used for offline database construction, and the text-side model is used for the online return. In the final online retrieval, the database data of the image side model will be searched after the text side model encodes Query, and the CLIP pre-training model guarantees the semantic correlation between images and texts. The model can draw the graphic pairs closer in vector space by pre-training on a large amount of visual data. Here we need to export the text-side model for online MetaSpore Serving inference. Since the retrieval scene is based on Chinese, the CLIP model supporting Chinese understanding is selected. The exported content includes the ONNX model used for online reasoning and Tokenizer, similar to the text search. MetaSpore Serving can load model reasoning through the exported content. Build library on Image search You need to download the Unsplash Lite library data and complete the construction according to the instructions. The whole process of database construction is described as follows: Preprocessing, specify the image directory, and then generate a more general JSOnline file for library construction; Build index, use OpenAI/Clip-Vit-BASE-Patch32 pre-training model to index the gallery, and output one document object for each line of index data; Push inverted (vector) and forward (document field) data to each component server. Like text search, after offline database construction, relevant data will be pushed to service components, called by online retrieval algorithm services to obtain relevant data. 1.2 Online Services The overall online service architecture diagram is as follows: https://preview.redd.it/jfsl8hdfez291.png?width=1280&format=png&auto=webp&s=a858e2304a0c93e78ba5429612ca08cbee69b35a Multi-mode search online service system supports application scenarios such as text search and text search. The whole online service consists of the following parts: Query preprocessing service: encapsulate preprocessing logic (including text/image, etc.) of pre-training model, and provide services through gRPC interface; Retrieval algorithm service: the whole algorithm processing link includes AB experiment tangent flow configuration, MetaSpore Serving call, vector recall, sorting, document summary, etc.; User entry service: provides a Web UI interface for users to debug and track down problems in the retrieval service. From a user request perspective, these services form invocation dependencies from back to front, so to build up a multimodal sample, you need to run each service from front to back first. Before doing this, remember to export the offline model, put it online and build the library first. This article will introduce the various parts of the online service system and make the whole service system step by step according to the following guidance. See the ReadME at the end of this article for more details. 1.2.1 Query preprocessing service Deep learning models tend to be based on tensors, but NLP/CV models often have a preprocessing part that translates raw text and images into tensors that deep learning models can accept. For example, NLP class models often have a pre-tokenizer to transform text data of string type into discrete tensor data. CV class models also have similar processing logic to complete the cropping, scaling, transformation, and other processing of input images through preprocessing. On the one hand, considering that this part of preprocessing logic is decoupled from tensor reasoning of the depth model, on the other hand, the reason of the depth model has an independent technical system based on ONNX, so MetaSpore disassembled this part of preprocessing logic. NLP pretreatment Tokenizer has been integrated into the Query pretreatment service. MetaSpore dismantlement with a relatively general convention. Users only need to provide preprocessing logic files to realize the loading and prediction interface and export the necessary data and configuration files loaded into the preprocessing service. Subsequent CV preprocessing logic will also be integrated in this manner. The preprocessing service currently provides the gRPC interface invocation externally and is dependent on the Query preprocessing (QP) module in the retrieval algorithm service. After the user request reaches the retrieval algorithm service, it will be forwarded to the service to complete the data preprocessing and continue the subsequent processing. The ReadMe provides details on how the preprocessing service is started, how the preprocessing model exported offline to cloud storage enters the service, and how to debug the service. To further improve the efficiency and stability of model reasoning, MetaSpore Serving implements a Python preprocessing submodule. So MetaSpore can provide gRPC services through user-specified preprocessor.py, complete Tokenizer or CV-related preprocessing in NLP, and translate requests into a Tensor that deep models can handle. Finally, the model inference is carried out by MetaSpore, Serving subsequent sub-modules. Presented here on the lot code: https://github.com/meta-soul/MetaSpore/compare/add\python\preprocessor 1.2.2 Retrieval algorithm services Retrieval algorithm service is the core of the whole online service system, which is responsible for the triage of experiments, the assembly of algorithm chains such as preprocessing, recall, sorting, and the invocation of dependent component services. The whole retrieval algorithm service is developed based on the Java Spring framework and supports multi-mode retrieval scenarios of text search and text search graph. Due to good internal abstraction and modular design, it has high flexibility and can be migrated to similar application scenarios at a low cost. Here’s a quick guide to configuring the environment to set up the retrieval algorithm service. See ReadME for more details: Install dependent components. Use Maven to install the online-Serving component Search for service configurations. Copy the template configuration file and replace the MongoDB, Milvus, and other configurations based on the development/production environment. Install and configure Consul. Consul allows you to synchronize the search service configuration in real-time, including cutting the flow of experiments, recall parameters, and sorting parameters. The project’s configuration file shows the current configuration parameters of text search and text search. The parameter modelName in the stage of pretreatment and recall is the corresponding model exported in offline processing. Start the service. Once the above configuration is complete, the retrieval service can be started from the entry script. Once the service is started, you can test it! For example, for a user with userId=10 who wants to query “How to renew ID card,” access the text search service. 1.2.3 User Entry Service Considering that the retrieval algorithm service is in the form of the API interface, it is difficult to locate and trace the problem, especially for the text search image scene can intuitively display the retrieval results to facilitate the iterative optimization of the retrieval algorithm. This paper provides a lightweight Web UI interface for text search and image search, a search input box, and results in a display page for users. Developed by Flask, the service can be easily integrated with other retrieval applications. The service calls the retrieval algorithm service and displays the returned results on the page. It’s also easy to install and start the service. Once you’re done, go to http://127.0.0.1:8090 to see if the search UI service is working correctly. See the ReadME at the end of this article for details. Multimodal system demonstration The multimodal retrieval service can be started when offline processing and online service environment configuration have been completed following the above instructions. Examples of textual searches are shown below. Enter the entry of the text search map application, enter “cat” first, and you can see that the first three digits of the returned result are cats: https://preview.redd.it/0n5nuyvhez291.png?width=1280&format=png&auto=webp&s=1e9c054f541d53381674b8d6001b4bf524506bd2 If you add a color constraint to “cat” to retrieve “black cat,” you can see that it does return a black cat: https://preview.redd.it/rzc0qjyjez291.png?width=1280&format=png&auto=webp&s=d5bcc503ef0fb3360c7740e60e295cf372dcad47 Further, strengthen the constraint on the search term, change it to “black cat on the bed,” and return results containing pictures of a black cat climbing on the bed:  https://preview.redd.it/c4b2q8olez291.png?width=1280&format=png&auto=webp&s=4f3817b0b9f07e1e68d1d4a8281702ba3834a00a The cat can still be found through the text search system after the color and scene modification in the above example. Conclusion The cutting-edge pre-training technology can bridge the semantic gap between different modes, and the HuggingFace community can greatly reduce the cost for developers to use the pre-training model. Combined with the technological ecology of MetaSpore online reasoning and online microservices provided by DMetaSpore, the pre-training model is no longer mere offline dabbling. Instead, it can truly achieve end-to-end implementation from cutting-edge technology to industrial scenarios, fully releasing the dividends of the pre-training large model. In the future, DMetaSoul will continue to improve and optimize the MetaSpore technology ecosystem: More automated and wider access to HuggingFace community ecology. MetaSpore will soon release a common model rollout mechanism to make HuggingFace ecologically accessible and will later integrate preprocessing services into online services. Multi-mode retrieval offline algorithm optimization. For multimodal retrieval scenarios, MetaSpore will continuously iteratively optimize offline algorithm components, including text recall/sort model, graphic recall/sort model, etc., to improve the accuracy and efficiency of the retrieval algorithm. For related code and reference documentation in this article, please visit: https://github.com/meta-soul/MetaSpore/tree/main/demo/multimodal/online Some images source: https://github.com/openai/CLIP/raw/main/CLIP.png https://www.sbert.net/examples/training/sts/README.html

#reddit #[p]mmml #deploy

[D] What's the endgame for AI labs that are spending billions on training generative models?

0

bendee983•This week

[D] What's the endgame for AI labs that are spending billions on training generative models?

Given the current craze around LLMs and generative models, frontier AI labs are burning through billions of dollars of VC funding to build GPU clusters, train models, give free access to their models, and get access to licensed data. But what is their game plan for when the excitement dies off and the market readjusts? There are a few challenges that make it difficult to create a profitable business model with current LLMs: The near-equal performance of all frontier models will commoditize the LLM market and force providers to compete over prices, slashing profit margins. Meanwhile, the training of new models remains extremely expensive. Quality training data is becoming increasingly expensive. You need subject matter experts to manually create data or review synthetic data. This in turn makes each iteration of model improvement even more expensive. Advances in open source and open weight models will probably take a huge part of the enterprise market of private models. Advances in on-device models and integration with OS might reduce demand for cloud-based models in the future. The fast update cycles of models gives AI companies a very short payback window to recoup the huge costs of training new models. What will be the endgame for labs such as Anthropic, Cohere, Mistral, Stability, etc. when funding dries up? Will they become more entrenched with big tech companies (e.g., OpenAI and Microsoft) to scale distribution? Will they find other business models? Will they die or be acquired (e.g., Inflection AI)? Thoughts?

#reddit #what's #endgame

0

1

Starks-Technology•This week

[D] What is your honest experience with reinforcement learning?

In my personal experience, SOTA RL algorithms simply don't work. I've tried working with reinforcement learning for over 5 years. I remember when Alpha Go defeated the world famous Go player, Lee Sedol, and everybody thought RL would take the ML community by storm. Yet, outside of toy problems, I've personally never found a practical use-case of RL. What is your experience with it? Aside from Ad recommendation systems and RLHF, are there legitimate use-cases of RL? Or, was it all hype? Edit: I know a lot about AI. I built NexusTrade, an AI-Powered automated investing tool that lets non-technical users create, update, and deploy their trading strategies. I’m not an idiot nor a noob; RL is just ridiculously hard. Edit 2: Since my comments are being downvoted, here is a link to my article that better describes my position. It's not that I don't understand RL. I released my open-source code and wrote a paper on it. It's the fact that it's EXTREMELY difficult to understand. Other deep learning algorithms like CNNs (including ResNets), RNNs (including GRUs and LSTMs), Transformers, and GANs are not hard to understand. These algorithms work and have practical use-cases outside of the lab. Traditional SOTA RL algorithms like PPO, DDPG, and TD3 are just very hard. You need to do a bunch of research to even implement a toy problem. In contrast, the decision transformer is something anybody can implement, and it seems to match or surpass the SOTA. You don't need two networks battling each other. You don't have to go through hell to debug your network. It just naturally learns the best set of actions in an auto-regressive manner. I also didn't mean to come off as arrogant or imply that RL is not worth learning. I just haven't seen any real-world, practical use-cases of it. I simply wanted to start a discussion, not claim that I know everything. Edit 3: There's a shockingly number of people calling me an idiot for not fully understanding RL. You guys are wayyy too comfortable calling people you disagree with names. News-flash, not everybody has a PhD in ML. My undergraduate degree is in biology. I self-taught myself the high-level maths to understand ML. I'm very passionate about the field; I just have VERY disappointing experiences with RL. Funny enough, there are very few people refuting my actual points. To summarize: Lack of real-world applications Extremely complex and inaccessible to 99% of the population Much harder than traditional DL algorithms like CNNs, RNNs, and GANs Sample inefficiency and instability Difficult to debug Better alternatives, such as the Decision Transformer Are these not legitimate criticisms? Is the purpose of this sub not to have discussions related to Machine Learning? To the few commenters that aren't calling me an idiot...thank you! Remember, it costs you nothing to be nice! Edit 4: Lots of people seem to agree that RL is over-hyped. Unfortunately those comments are downvoted. To clear up some things: We've invested HEAVILY into reinforcement learning. All we got from this investment is a robot that can be super-human at (some) video games. AlphaFold did not use any reinforcement learning. SpaceX doesn't either. I concede that it can be useful for robotics, but still argue that it's use-cases outside the lab are extremely limited. If you're stumbling on this thread and curious about an RL alternative, check out the Decision Transformer. It can be used in any situation that a traditional RL algorithm can be used. Final Edit: To those who contributed more recently, thank you for the thoughtful discussion! From what I learned, model-based models like Dreamer and IRIS MIGHT have a future. But everybody who has actually used model-free models like DDPG unanimously agree that they suck and don’t work.

#reddit #your #honest

How a Small Startup in Asia Secured a Contract with the US Department of Homeland Security

0

1

Royal_Rest8409•This week

How a Small Startup in Asia Secured a Contract with the US Department of Homeland Security

Uzair Javaid, a Ph.D. with a passion for data privacy, co-founded Betterdata to tackle one of AI's most pressing challenges: protecting privacy while enabling innovation. Recently, Betterdata secured a lucrative contract with the US Department of Homeland Security, 1 of only 4 companies worldwide to do so and the only one in Asia. Here's how he did it: The Story So what's your story? I grew up in Peshawar, Pakistan, excelling in coding despite studying electrical engineering. Inspired by my professors, I set my sights on studying abroad and eventually earned a Ph.D. scholarship at NUS Singapore, specializing in data security and privacy. During my research, I ethically hacked Ethereum and published 15 papers—three times the requirement. While wrapping up my Ph.D., I explored startup ideas and joined Entrepreneur First, where I met Kevin Yee. With his expertise in generative models and mine in privacy, we founded Betterdata. Now, nearly three years in, we’ve secured a major contract with the U.S. Department of Homeland Security—one of only four companies globally and the only one from Asia. The Startup In a nutshell, what does your startup do? Betterdata is a startup that uses AI and synthetic data generation to address two major challenges: data privacy and the scarcity of high-quality data for training AI models. By leveraging generative models and privacy-enhancing technologies, Betterdata enables businesses, such as banks, to use customer data without breaching privacy regulations. The platform trains AI on real data, learns its patterns, and generates synthetic data that mimics the real thing without containing any personal or sensitive information. This allows companies to innovate and develop AI solutions safely and ethically, all while tackling the growing need for diverse, high-quality data in AI development. How did you conduct ideation and validation for your startup? The initial idea for Betterdata came from personal experience. During my Ph.D., I ethically hacked Ethereum’s blockchain, exposing flaws in encryption-based data sharing. This led me to explore AI-driven deep synthesis technology—similar to deepfakes but for structured data privacy. With GDPR impacting 28M+ businesses, I saw a massive opportunity to help enterprises securely share data while staying compliant. To validate the idea, I spoke to 50 potential customers—a number that strikes the right balance. Some say 100, but that’s impractical for early-stage founders. At 50, patterns emerge: if 3 out of 10 mention the same problem, and this repeats across 50, you have 10–15 strong signals, making it a solid foundation for an MVP. Instead of outbound sales, which I dislike, we used three key methods: Account-Based Marketing (ABM)—targeting technically savvy users with solutions for niche problems, like scaling synthetic data for banks. Targeted Content Marketing—regular customer conversations shaped our thought leadership and outreach. Raising Awareness Through Partnerships—collaborating with NUS, Singapore’s PDPC, and Plug and Play to build credibility and educate the market. These strategies attracted serious customers willing to pay, guiding Betterdata’s product development and market fit. How did you approach the initial building and ongoing product development? In the early stages, we built synthetic data generation algorithms and a basic UI for proof-of-concept, using open-source datasets to engage with banks. We quickly learned that banks wouldn't share actual customer data due to privacy concerns, so we had to conduct on-site installations and gather feedback to refine our MVP. Through continuous consultation with customers, we discovered real enterprise data posed challenges, such as missing values, which led us to adapt our prototype accordingly. This iterative approach of listening to customer feedback and observing their usage allowed us to improve our product, enhance UX, and address unmet needs while building trust and loyalty. Working closely with our customers also gives us a data advantage. Our solution’s effectiveness depends on customer data, which we can't fully access, but bridging this knowledge gap gives us a competitive edge. The more customers we test on, the more our algorithms adapt to diverse use cases, making it harder for competitors to replicate our insights. My approach to iteration is simple: focus solely on customer feedback and ignore external noise like trends or advice. The key question for the team is: which customer is asking for this feature or solution? As long as there's a clear answer, we move forward. External influences, such as AI hype, often bring more confusion than clarity. True long-term success comes from solving real customer problems, not chasing trends. Customers may not always know exactly what they want, but they understand their problems. Our job is to identify these problems and solve them in innovative ways. While customers may suggest specific features, we stay focused on solving the core issue rather than just fulfilling their exact requests. The idea aligns with the quote often attributed to Henry Ford: "If I asked people what they wanted, they would have said faster horses." The key is understanding their problems, not just taking requests at face value. How do you assess product-market fit? To assess product-market fit, we track two key metrics: Customers' Willingness to Pay: We measure both the quantity and quality of meetings with potential customers. A high number of meetings with key decision-makers signals genuine interest. At Betterdata, we focused on getting meetings with people in banks and large enterprises to gauge our product's resonance with the target market. How Much Customers Are Willing to Pay: We monitor the price customers are willing to pay, especially in the early stages. For us, large enterprises, like banks, were willing to pay a premium for our synthetic data platform due to the growing need for privacy tech. This feedback guided our product refinement and scaling strategy. By focusing on these metrics, we refined our product and positioned it for scaling. What is your business model? We employ a structured, phase-driven approach for out business model, as a B2B startup. I initially struggled with focusing on the core value proposition in sales, often becoming overly educational. Eventually, we developed a product roadmap with models that allowed us to match customer needs to specific offerings and justify our pricing. Our pricing structure includes project-based pilots and annual contracts for successful deployments. At Betterdata, our customer engagement unfolds across three phases: Phase 1: Trial and Benchmarking \- We start with outreach and use open-source datasets to showcase results, offering customers a trial period to evaluate the solution. Phase 2: Pilot or PoC \- After positive trial results, we conduct a PoC or pilot using the customer’s private data, with the understanding that successful pilots lead to an annual contract. Phase 3: Multi-Year Contracts \- Following a successful pilot, we transition to long-term commercial contracts, focusing on multi-year agreements to ensure stability and ongoing partnerships. How do you do marketing for your brand? We take a non-conventional approach to marketing, focusing on answering one key question: Which customers are willing to pay, and how much? This drives our messaging to show how our solution meets their needs. Our strategy centers around two main components: Building a network of lead magnets \- These are influential figures like senior advisors, thought leaders, and strategic partners. Engaging with institutions like IMDA, SUTD, and investors like Plug and Play helps us gain access to the right people and foster warm introductions, which shorten our sales cycle and ensure we’re reaching the right audience. Thought leadership \- We build our brand through customer traction, technology evidence, and regulatory guidelines. This helps us establish credibility in the market and position ourselves as trusted leaders in our field. This holistic approach has enabled us to navigate diverse market conditions in Asia and grow our B2B relationships. By focusing on these areas, we drive business growth and establish strong trust with stakeholders. What's your advice for fundraising? Here are my key takeaways for other founders when it comes to fundraising: Fundraise When You Don’t Need To We closed our seed round in April 2023, a time when we weren't actively raising. Founders should always be in fundraising mode, even when they're not immediately in need of capital. Don’t wait until you have only a few months of runway left. Keep the pipeline open and build relationships. When the timing is right, execution becomes much easier. For us, our investment came through a combination of referrals and inbound interest. Even our lead investor initially rejected us, but after re-engaging, things eventually fell into place. It’s crucial to stay humble, treat everyone with respect, and maintain those relationships for when the time is right. Be Mindful of How You Present Information When fundraising, how you present information matters a lot. We created a comprehensive, easily digestible investment memo, hosted on Notion, which included everything an investor might need—problem, solution, market, team, risks, opportunities, and data. The goal was for investors to be able to get the full picture within 30 minutes without chasing down extra details. We also focused on making our financial model clear and meaningful, even though a 5-year forecast might be overkill at the seed stage. The key was clarity and conciseness, and making it as easy as possible for investors to understand the opportunity. I learned that brevity and simplicity are often the best ways to make a memorable impact. For the pitch itself, keep it simple and focus on 4 things: problem, solution, team, and market. If you can summarize each of these clearly and concisely, you’ll have a compelling pitch. Later on, you can expand into market segments, traction, and other metrics, but for seed-stage, focus on those four areas, and make sure you’re strong in at least three of them. If you do, you'll have a compelling case. How do you run things day-to-day? i.e what's your operational workflow and team structure? Here's an overview of our team structure and process: Internally: Our team is divided into two main areas: backend (internal team) and frontend (market-facing team). There's no formal hierarchy within the backend team. We all operate as equals, defining our goals based on what needs to be developed, assigning tasks, and meeting weekly to share updates and review progress. The focus is on full ownership of tasks and accountability for getting things done. I also contribute to product development, identifying challenges and clearing obstacles to help the team move forward. Backend Team: We approach tasks based on the scope defined by customers, with no blame or hierarchy. It's like a sports team—sometimes someone excels, and other times they struggle, but we support each other and move forward together. Everyone has the creative freedom to work in the way that suits them best, but we establish regular meetings and check-ins to ensure alignment and progress. Frontend Team: For the market-facing side, we implement a hierarchy because the market expects this structure. If I present myself as "CEO," it signals authority and credibility. This distinction affects how we communicate with the market and how we build our brand. The frontend team is split into four main areas: Business Product (Software Engineering) Machine Learning Engineering R&D The C-suite sits at the top, followed by team leads, and then the executors. We distill market expectations into actionable tasks, ensuring that everyone is clear on their role and responsibilities. Process: We start by receiving market expectations and defining tasks based on them. Tasks are assigned to relevant teams, and execution happens with no communication barriers between team members. This ensures seamless collaboration and focused execution. The main goal is always effectiveness—getting things done efficiently while maintaining flexibility in how individuals approach their work. In both teams, there's an emphasis on accountability, collaboration, and clear communication, but the structure varies according to the nature of the work and external expectations.

#reddit #small #startup

Started a content marketing agency 6 years ago - $0 to $5,974,324 (2023 update)

0

1

mr_t_forhire•This week

Started a content marketing agency 6 years ago - $0 to $5,974,324 (2023 update)

Hey friends, My name is Tyler and for the past 6 years, I’ve been documenting my experience building a content marketing agency called Optimist. Year 1 - 0 to $500k ARR Year 2 - $500k to $1MM ARR Year 3 - $1MM ARR to $1.5MM(ish) ARR Year 4 - $3,333,686 Revenue Year 5 - $4,539,659 Revenue How Optimist Works First, an overview/recap of the Optimist business model: We operate as a “collective” of full time/professional freelancers Everyone aside from me is a contractor Entirely remote/distributed team Each freelancer earns $65-85/hour Clients pay us a flat monthly fee for full-service content marketing (research, strategy, writing, editing, design/photography, reporting and analytics, targeted linkbuilding, and more) We recently introduced hourly engagements for clients who fit our model but have some existing in-house support Packages range in price from $10-20k/mo We offer profit share to everyone on our core team as a way to give everyone ownership in the company In 2022, we posted $1,434,665 in revenue. It was our highest revenue year to date and brings our lifetime total to $5,974,324. Here’s our monthly revenue from January 2017 to December of 2022. But, like every year, it was a mix of ups and downs. Here’s my dispatch for 2023. — Running a business is like spilling a drink. It starts as a small and simple thing. But, if you don’t clean it up, the spill will spread and grow — taking up more space, seeping into every crack. There’s always something you could be doing. Marketing you could be working on. Pitches you could be making. Networking you could be doing. Client work you could help with. It can be all-consuming. And it will be — if you don’t clean up the spill. I realized this year that I had no containment for the spill that I created. Running an agency was spilling over into nearly every moment of my life. When I wasn’t working, I was thinking about work. When I wasn’t thinking about work, I was dreaming about it. Over the years, I’ve shared about a lot of my personal feelings and experience as an entrepreneur. And I also discussed my reckoning with the limitations of running the business we’ve built. My acceptance that it was an airplane but not a rocket. And my plan to try to compartmentalize the agency to make room in my life for other things — new business ideas, new revenue streams, and maybe some non-income-producing activity. 🤷 What I found in 2022 was that the business wasn’t quite ready for me to make that move. It was still sucking up too much of my time and attention. There were still too many gaps to fill and I was the one who was often filling them. So what do you do? Ultimately you have two choices on the table anytime you run a business and it’s not going the way you want it: Walk away Turn the ship — slowly For a huge number of reasons (personal, professional, financial, etc), walking away from Optimist was not really even an option or the right move for me. But it did feel like things needed to change. I needed to keep turning the ship to get it to the place where it fit into my life — instead of my life fitting around the business. This means 2022 was a year of transition for the agency. (Again?) Refocusing on Profit Some money is better than no money. Right? Oddly, this was one of the questions I found myself asking in 2022. Over the years, we’ve been fortunate to have many clients who have stuck with us a long time. In some cases, we’ve had clients work with us for 2, 3, or even 4 years. (That’s over half of our existence!) But, things have gotten more expensive — we’ve all felt it. We’ve had to increase pay to remain competitive for top talent. Software costs have gone up. It’s eaten into our margin. Because of our increasing costs and evolving scope, many of our best, most loyal clients were our least profitable. In fact, many were barely profitable — if at all. We’ve tried to combat that by increasing rates on new, incoming clients to reflect our new costs and try to make up for shrinking margin on long-term clients. But we didn’t have a good strategy in place for updating pricing for current clients. And it bit us in the ass. Subsidizing lower-profit, long-term clients with new, higher-margin clients ultimately didn’t work out. Our margins continued to dwindle and some months we were barely breaking even while posting six-figures of monthly revenue. 2022 was our highest revenue year but one of our least profitable. It only left one option. We had to raise rates on some of our long-term clients. But, of course, raising rates on a great, long-term client can be delicate. You’ve built a relationship with these people over the years and you’re setting yourself up for an ultimatum — are you more valuable to the client or is the client more valuable to you? Who will blink first? We offered all of these clients the opportunity to move to updated pricing. Unfortunately, some of them weren’t on board. Again, we had 2 options: Keep them at a low/no profit rate Let them churn It seems intuitive that having a low-profit client is better than having no client. But we’ve learned an important lesson many times over the years. Our business doesn’t scale infinitely and we can only handle so many clients at a time. That means that low-profit clients are actually costing us money in some cases. Say our average client generates $2,500 per month in profit — $30,000 per year. If one of our clients is only generating $500/mo in profit, working with them means missing out on bringing on a more profitable client (assuming our team is currently at capacity). Instead of $30,000/year, we’re only making $6,000. Keeping that client costs us $24,000. That’s called opportunity cost. So it’s clear: We had to let these clients churn. We decided to churn about 25% of our existing clients. On paper, the math made sense. And we had a pretty consistent flow of new opportunities coming our way. At the time, it felt like a no-brainer decision. And I felt confident that we could quickly replace these low-profit clients with higher-margin ones. I was wrong. Eating Shit Right after we initiated proactively churning some of our clients, other clients — ones we planned to keep — gave us notice that they were planning to end the engagement. Ouch. Fuck. We went from a 25% planned drop in revenue to a nearly 40% cliff staring us right in the face. Then things got even worse. Around Q3 of this year, talk of recession and layoffs really started to intensify. We work primarily with tech companies and startups. And these were the areas most heavily impacted by the economic news. Venture funding was drying up. Our leads started to slow down. This put us in a tough position. Looking back now, I think it’s clear that I made the wrong decision. We went about this process in the wrong way. The reality sinks in when you consider the imbalance between losing a client and gaining a client. It takes 30 days for someone to fire us. It’s a light switch. But it could take 1-3 months to qualify, close, and onboard a new client. We have lots of upfront work, research, and planning that goes into the process. We have to learn a new brand voice, tone, and style. It’s a marathon. So, for every client we “trade”, there’s a lapse in revenue and work. This means that, in retrospect, I would probably have made this transition using some kind of staggered schedule rather than a cut-and-dry approach. We could have gradually off-boarded clients when we had more definitive work to replace them. I was too confident. But that’s a lesson I had to learn the hard way. Rebuilding & Resetting Most of the voluntary and involuntary churn happened toward the end of 2022. So we’re still dealing with the fall out. Right now, it feels like a period of rebuilding. We didn’t quite lose 50% of our revenue, but we definitely saw a big hit heading into 2023. To be transparent: It sucks. It feels like a gigantic mistake that I made which set us back significantly from our previous high point. I acted rashly and it cost us a lot of money — at least on the surface. But I remind myself of the situation we were in previously. Nearly twice the revenue but struggling to maintain profitability. Would it have been better to try to slowly fix that situation and battle through months of loss or barely-break-even profits? Or was ripping off the bandaid the right move after all? I’m an optimist. (Heh, heh) Plus, I know that spiraling over past decisions won’t change them or help me move forward. So I’m choosing to look at this as an opportunity — to rebuild, reset, and refocus the company. I get to take all of the tough lessons I’ve learned over the last 6 years and apply them to build the company in a way that better aligns with our new and current goals. It’s not quite a fresh, clean start, but by parting ways with some of our oldest clients, we’ve eliminated some of the “debt” that’s accumulated over the years. We get a chance to fully realize the new positioning that we rolled out last year. Many of those long-term clients who churned had a scope of work or engagement structure that didn’t fit with our new positioning and focus. So, by losing them, we’re able to completely close up shop on the SOWs that no longer align with the future version of Optimist. Our smaller roster of clients is a better fit for that future. My job is to protect that positioning by ensuring that while we’re rebuilding our new roster of clients we don’t get desperate. We maintain the qualifications we set out for future clients and only take on work that fits. How’s that for seeing the upside? Some other upside from the situation is that we got an opportunity to ask for candid feedback from clients who were leaving. We asked for insight about their decision, what factors they considered, how they perceived us, and the value of our work. Some of the reasons clients left were obvious and possibly unavoidable. Things like budget cuts, insourcing, and uncertainty about the economy all played at least some part of these decisions. But, reading between the lines, where was one key insight that really struck me. It’s one of those, “oh, yeah — duh — I already knew that,” things that can be difficult to learn and easy to forget…. We’re in the Relationship Business (Plan Accordingly) For all of our focus on things like rankings, keywords, content, conversions, and a buffet of relevant metrics, it can be easy to lose the forest for the trees. Yes, the work itself matters. Yes, the outcomes — the metrics — matter. But sometimes the relationship matters more. When you’re running an agency, you can live or die by someone just liking you. Admittedly, this feels totally unfair. It opens up all kinds of dilemmas, frustration, opportunity for bias and prejudice, and other general messiness. But it’s the real world. If a client doesn’t enjoy working with us — even if for purely personal reasons — they could easily have the power to end of engagement, regardless of how well we did our actual job. We found some evidence of this in the offboarding conversations we had with clients. In some cases, we had clients who we had driven triple- and quadruple-digital growth. Our work was clearly moving the needle and generating positive ROI and we had the data to prove it. But they decided to “take things in another direction” regardless. And when we asked about why they made the decision, it was clear that it was more about the working relationship than anything we could have improved about the service itself. The inverse is also often true. Our best clients have lasting relationships with our team. The work is important — and they want results. But even if things aren’t quite going according to plan, they’re patient and quick to forgive. Those relationships feel solid — unshakeable. Many of these folks move onto new roles or new companies and quickly look for an opportunity to work with us again. On both sides, relationships are often more important than the work itself. We’ve already established that we’re not building a business that will scale in a massive way. Optimist will always be a small, boutique service firm. We don’t need 100 new leads per month We need a small, steady roster of clients who are a great fit for the work we do and the value we create. We want them to stick around. We want to be their long-term partner. I’m not built for churn-and-burn agency life. And neither is the business. When I look at things through this lens, I realize how much I can cut from our overall business strategy. We don’t need an ultra-sophisticated, multi-channel marketing strategy. We just need strong relationships — enough of them to make our business work. There are a few key things we can take away from this as a matter of business strategy: Put most of our effort into building and strengthening relationships with our existing clients Be intentional about establishing a strong relationship with new clients as part of onboarding Focus on relationships as the main driver of future business development Embracing Reality: Theory vs Practice Okay, so with the big learnings out the way, I want to pivot into another key lesson from 2022. It’s the importance of understanding theory vs practice — specifically when it comes to thinking about time, work, and life. It all started when I was considering how to best structure my days and weeks around running Optimist, my other ventures, and my life goals outside of work. Over the years, I’ve dabbled in many different ways to block time and find focus — to compartmentalize all of the things that are spinning and need my attention. As I mapped this out, I realized that I often tried to spread myself too thin throughout the week. Not just that I was trying to do too much but that I was spreading that work into too many small chunks rather than carving out time for focus. In theory, 5 hours is 5 hours. If you have 5 hours of work to get done, you just fit into your schedule whenever you have an open time slot. In reality, a single 5-hour block of work is 10x more productive and satisfying than 10, 30-minute blocks of work spread out across the week. In part, this is because of context switching. Turning your focus from one thing to another thing takes time. Achieving flow and focus takes time. And the more you jump from one project to another, the more time you “lose” to switching. This is insightful for me both in the context of work and planning my day, but also thinking about my life outside of Optimist. One of my personal goals is to put a finite limit on my work time and give myself more freedom. I can structure that in many different ways. Is it better to work 5 days a week but log off 1 hour early each day? Or should I try to fit more hours into each workday so I can take a full day off? Of course, it’s the latter. Both because of the cost of context switching and spreading work into more, smaller chunks — but also because of the remainder that I end up with when I’m done working. A single extra hour in my day probably means nothing. Maybe I can binge-watch one more episode of a new show or do a few extra chores around the house. But it doesn’t significantly improve my life or help me find greater balance. Most things I want to do outside of work can’t fit into a single extra hour. A full day off from work unlocks many more options. I can take the day to go hiking or biking. I can spend the day with my wife, planning or playing a game. Or I can push it up against the weekend and take a 3-day trip. It gives me more of the freedom and balance that I ultimately want. So this has become a guiding principle for how I structure my schedule. I want to: Minimize context switching Maximize focused time for work and for non-work The idea of embracing reality also bleeds into some of the shifts in business strategy that I mentioned above. In theory, any time spent on marketing will have a positive impact on the company. In reality, focusing more on relationships than blasting tweets into the ether is much more likely to drive the kind of growth and stability that we’re seeking. As I think about 2023, I think this is a recurring theme. It manifests in many ways. Companies are making budget cuts and tough decisions about focus and strategy. Most of us are looking for ways to rein in the excess and have greater impact with a bit less time and money. We can’t do everything. We can’t even do most things. So our #1 priority should be to understand the reality of our time and our effort to make the most of every moment (in both work and leisure). That means thinking deeply about our strengths and our limitations. Being practical, even if it feels like sacrifice. Update on Other Businesses Finally, I want to close up by sharing a bit about my ventures outside of Optimist. I shared last year how I planned to shift some of my (finite) time and attention to new ventures and opportunities. And, while I didn’t get to devote as much as I hoped to these new pursuits, they weren’t totally in vain. I made progress across the board on all of the items I laid out in my post. Here’s what happened: Juice: The first Optimist spin-out agency At the end of 2021, we launched our first new service business based on demand from Optimist clients. Focused entirely on building links for SEO, we called the agency Juice. Overall, we made strong progress toward turning this into a legitimate standalone business in 2022. Relying mostly on existing Optimist clients and a few word-of-mouth opportunities (no other marketing), we built a team and set up a decent workflow and operations. There’s still many kinks and challenges that we’re working through on this front. All told, Juice posted almost $100,000 in revenue in our first full year. Monetizing the community I started 2022 with a focus on figuring out how to monetize our free community, Top of the Funnel. Originally, my plan was to sell sponsorships as the main revenue driver. And that option is still on the table. But, this year, I pivoted to selling paid content and subscriptions. We launched a paid tier for content and SEO entrepreneurs where I share more of my lessons, workflows, and ideas for building and running a freelance or agency business. It’s gained some initial traction — we reached \~$1,000 MRR from paid subscriptions. In total, our community revenue for 2022 was about $2,500. In 2023, I’m hoping to turn this into a $30,000 - $50,000 revenue opportunity. Right now, we’re on track for \~$15,000. Agency partnerships and referrals In 2022, we also got more serious about referring leads to other agencies. Any opportunity that was not a fit for Optimist or we didn’t have capacity to take on, we’d try to connect with another partner. Transparently, we struggled to operationalize this as effectively as I would have liked. In part, this was driven by my lack of focus here. With the other challenges throughout the year, I wasn’t able to dedicate as much time as I’d like to setting goals and putting workflows into place. But it wasn’t a total bust. We referred out several dozen potential clients to partner agencies. Of those, a handful ended up converting into sales — and referral commission. In total, we generated about $10,000 in revenue from referrals. I still see this as a huge opportunity for us to unlock in 2023. Affiliate websites Lastly, I mentioned spending some time on my new and existing affiliate sites as another big business opportunity in 2022. This ultimately fell to the bottom of my list and didn’t get nearly the attention I wanted. But I did get a chance to spend a few weeks throughout the year building this income stream. For 2022, I generated just under $2,000 in revenue from affiliate content. My wife has graciously agreed to dedicate some of her time and talent to these projects. So, for 2023, I think this will become a bit of a family venture. I’m hoping to build a solid and consistent workflow, expand the team, and develop a more solid business strategy. Postscript — AI, SEO, OMG As I’m writing this, much of my world is in upheaval. If you’re not in this space (and/or have possibly been living under a rock), the release of ChatGPT in late 2022 has sparked an arms race between Google, Bing, OpenAI, and many other players. The short overview: AI is likely to fundamentally change the way internet search works. This has huge impact on almost all of the work that I do and the businesses that I run. Much of our focus is on SEO and understanding the current Google algorithm, how to generate traffic for clients, and how to drive traffic to our sites and projects. That may all change — very rapidly. This means we’re standing at a very interesting point in time. On the one hand, it’s scary as hell. There’s a non-zero chance that this will fundamentally shift — possibly upturn — our core business model at Optimist. It could dramatically change how we work and/or reduce demand for our core services. No bueno. But it’s also an opportunity (there’s the optimist in me, again). I certainly see a world where we can become leaders in this new frontier. We can pivot, adjust, and capitalize on a now-unknown version of SEO that’s focused on understanding and optimizing for AI-as-search. With that, we may also be able to help others — say, those in our community? — also navigate this tumultuous time. See? It’s an opportunity. I wish I had the answers right now. But, it’s still a time of uncertainty. I just know that there’s a lot of change happening and I want to be in front of it rather than trying to play catch up. Wish me luck. — Alright friends — that's my update for 2023! I’ve always appreciated sharing these updates with the Reddit community, getting feedback, being asked tough questions, and even battling it out with some of my haters (hey!! 👋) As usual, I’m going to pop in throughout the next few days to respond to comments or answer questions. Feel free to share thoughts, ideas, and brutal takedowns in the comments. If you're interested in following the Optimist journey and the other projects I'm working on in 2023, you can follow me on Twitter. Cheers, Tyler P.S. - If you're running or launching a freelance or agency business and looking for help figuring it out, please DM me. Our subscription community, Middle of the Funnel, was created to provide feedback, lessons, and resources for other entrepreneurs in this space.

#reddit #started #content

I got fired due to automation — lessons learned. Two-month overview.

0

1

WebsterPepster•This week

I got fired due to automation — lessons learned. Two-month overview.

UPD: Guys, I'm not promoting myself as some of the redditors decided. That's why to deal with contradictions I'll do next things: make additional post with short review and description of the general tools and processes you could apply. help only those who have already written me. So I won't answer on new offers or DMs. As mentioned, damn robots have taken my job. PRE-HISTORY During Covid times, I found myself without my offline job, and since I was interested in marketing and SMM, I began searching for a job there. Completed free Google and Udemy courses and finally landed my first SMM manager position with a business owner. He had several projects so, finally, I started managing three Twitter accounts, two Facebook accs, two IGs, and one TikTok. I handled posting, content editing and responding routine, while freelancers usually took care of video creation for IG and TT. THE STORY ITSELF Things took a turn for the worse in April when my employer introduced ChatGPT and Midjourney, tools I was already using. The owner insisted on integrating them into the workflow, and my wages took a 20% hit. I thought I could roll with it, but it was just the beginning. By midsummer, the owner implemented second-layer AI tools like Visla, Pictory, and Woxo for video (bye freelancers, lol), as well as TweetHunter, Jasper, and Perplexity for content. Midjourney and Firefly joined for image generation. All together, my paycheck was slashed by 50%. Finally, at the end of October, my boss told me he automated stuff with Zapier, cutting costs that way. Additionally, he adopted MarketOwl, autoposting tool for Twitter, and SocialBee for Facebook. He stated that he didn’t need me, as by now he could manage the social media accounts himself. I feel so pissed then and even thought that there's no point in searching for similar jobs. HOW I SPENT TWO MONTHS Well, for the first two weeks, I did nothing but being miserable, drinking and staring at the wall. My gf said it's unbearable and threatened to leave if I not pull myself together. It was not the final push, but definitely made me rethink things. So I decided to learn more about the capabilities of these automation covers and eventually became an AI adviser for small businesses. It's ironic that now I sometimes earn money advising on how to optimize marketing, possibly contributing to other people's job loss. FINAL THOUGHTS I am fully aware of the instability of such a job and have invested my last savings in taking an online marketing course at Columbia to gain more marketing experience and got something more stable afterwards. Message for mods: I'm not promoting myself or anything mentioned here; just sharing the experience that someone might find helpful.

#reddit #fired #automation

Started a content marketing agency 8 years ago - $0 to $7,863,052 (2025 update)

0

0.882

mr_t_forhire•This week

Started a content marketing agency 8 years ago - $0 to $7,863,052 (2025 update)

Hey friends, My name is Tyler and for the past 8 years, I’ve been documenting my experience building a content marketing agency called Optimist. Year 1 — 0 to $500k ARR Year 2 — $500k to $1MM ARR Year 3 — $1MM ARR to $1.5MM(ish) ARR Year 4 — $3,333,686 Revenue Year 5 — $4,539,659 Revenue Year 6 — $5,974,324 Revenue Year 7 - $6,815,503 Revenue (Edit: Seems like links are banned now. You can check my post history for all of my previous updates with lessons and learnings.) How Optimist Works First, an overview/recap of the Optimist business model: We operate as a “collective” of full time/professional freelancers Everyone aside from me is a contractor Entirely remote/distributed team We pay freelancers a flat fee for most work, working out to roughly $65-100/hour. Clients pay us a flat monthly fee for full-service content marketing (research, strategy, writing, editing, design/photography, reporting and analytics, targeted linkbuilding, and more)\ Packages range in price from \~$10-20k/mo \This is something we are revisiting now* The Financials In 2024, we posted $1,032,035.34 in revenue. This brings our lifetime revenue to $7,863,052. Here’s our monthly revenue from January 2017 to December of 2024. (Edit: Seems like I'm not allowed to link to the chart.) The good news: Revenue is up 23% YoY. EBITDA in Q4 trending up 1-2 points. We hosted our first retreat in 4 years, going to Ireland with about half the team. The bad news: Our revenue is still historically low. At $1MM for the year, we’re down about 33% from our previous years over $1.5MM. Revenue has been rocky. It doesn’t feel like we’ve really “recovered” from the bumps last year. The trend doesn’t really look great. Even though, anecdotally, it feels like we are moving in a good direction. EBITDA is still hovering at around 7%. Would love to get that closer to 20%. (For those who may ask: I’m calculating EBITDA after paying taxes and W2 portion of my income.) — Almost every year, my update starts the same way: This has been a year of growth and change. Both for my business—and me personally. 2024 was no different. I guess that tells you something about entrepreneurship. It’s a lot more like sailing a ship than driving a car. You’re constantly adapting, tides are shifting, and any blip of calm is usually just a moment before the next storm. As with past years, there’s a lot to unpack from the last 12 months. Here we go again. Everything is Burning In the last 2 years, everything has turned upside down in the world of content and SEO. Back in 2020, we made a big decision to re-position the agency. (See post history) We decided to narrow our focus to our most successful, profitable, and consistent segment of clients and re-work our entire operation to focus on serving them. We defined our ICP as: \~Series A ($10mm+ funding) with 6-12 months runway to scale organic as a channel Product-led company with “simple” sales cycle involving fewer stakeholders Demonstrable opportunity to use SEO to drive business growth Our services: Content focused on growing organic search (SEO) Full-service engagements that included research, planning, writing, design, reporting And our engagement structure: Engaged directly with an executive; ownership over strategy and day-to-day execution 1-2 points of contact or stakeholders Strategic partner that drives business growth (not a service vendor who makes content) Most importantly, we decided that we were no longer going to offer a broader range of content that we used to sell. That included everything from thought leadership content to case studies and ebooks. We doubled-down on “SEO content” for product-led SaaS companies. And this worked phenomenally for us. We started bringing on more clients than ever. We developed a lot of internal system and processes that helped us scale and take on more work than we’ve ever had and drive great outcomes for our ideal clients. But in 2023 and 2024, things started going awry. One big change, of course, was the rise of AI. Many companies and executives (and writers) feel that AI can write content just as well as an agency like ours. That made it a lot harder to sell a $10,000 per month engagement when they feel like the bulk of the work could be “done for free.” (Lots of thoughts on this if you want my opinions.) But it wasn’t just that. Google also started tinkering with their algorithm, introducing new features like AI Overviews, and generally changing the rules of the game. This created 3 big shifts in our world: The perceived value of content (especially “SEO content”) dropped dramatically in many people’s minds because of AI’s writing capabilities SEO became less predictable as a source of traffic and revenue It’s harder than ever for startups and smaller companies to rank for valuable keywords (let alone generate any meaningful traffic or revenue from them) The effect? The middle of the content market has hollowed out. People—like us—providing good, human-crafted content aimed on driving SEO growth saw a dramatic decline in demand. We felt it all year. Fewer and fewer leads. The leads we did see usually scoffed at our prices. They were indexing us against the cost of content mills and mass-produced AI articles. It was a time of soul-searching and looking for a way forward. I spent the first half of the year convinced that the only way to survive was to run toward the fire. We have to build our own AI workflows. We have to cut our rates internally. We have to get faster and cheaper to stay competitive with the agencies offering the same number of deliverables for a fraction of our rates. It’s the only way forward. But then I asked myself a question… Is this the game I actually want to play? As an entrepreneur, do I want to run a business where I’m competing mostly on price and efficiency rather than quality and value? Do I want to hop into a race toward cheaper and cheaper content? Do I want to help people chase a dwindling amount of organic traffic that’s shrinking in value? No. That’s not the game I want to play. That’s not a business I want to run. I don’t want to be in the content mill business. So I decided to turn the wheel—again. Repositioning Part II: Electric Boogaloo What do you do when the whole world shifts around you and the things that used to work aren’t working anymore? You pivot. You re-position the business and move in another direction. So that’s what we decided to do. Again. There was only one problem: I honestly wasn’t sure what opportunities existed in the content marketing industry outside of what we were already doing. We lived in a little echo chamber of startups and SEO. It felt like the whole market was on fire and I had fight through the smoke to find an escape hatch. So I started making calls. Good ol’ fashioned market research. I reached out to a few dozen marketing and content leaders at a bunch of different companies. I got on the phone and just asked lots of questions about their content programs, their goals, and their pain points. I wanted to understand what was happening in the market and how we could be valuable. And, luckily, this process really paid off. I learned a lot about the fragmentation happening across content and how views were shifting. I noticed key trends and how our old target market really wasn’t buying what we were selling. Startups and small companies are no longer willing to invest in an agency like ours. If they were doing content and SEO at all, they were focused entirely on using AI to scale output and minimize costs. VC money is still scarce and venture-backed companies are more focused on profitability than pure growth and raising another round. Larger companies (\~500+ employees) are doing more content than ever and drowning in content production. They want to focus on strategy but can barely tread water keeping up with content requests from sales, demand gen, the CEO, and everyone else. Many of the companies still investing in content are looking at channels and formats outside of SEO. Things like thought leadership, data reports, interview-driven content, and more. They see it as a way to stand out from the crowd of “bland SEO content.” Content needs are constantly in flux. They range from data reports and blog posts to product one-pagers. The idea of a fixed-scope retainer is a total mismatch for the needs of most companies. All of this led to the logical conclusion: We were talking to the wrong people about the wrong things\.\ Many companies came to one of two logical conclusions: SEO is a risky bet, so it’s gotta be a moonshot—super-low cost with a possibility for a big upside (i.e., use AI to crank out lots of content. If it works, great. If it doesn’t, then at least we aren’t out much money.) SEO is a risky bet, so we should diversify into other strategies and channels to drive growth (i.e., shift our budget from SEO and keyword-focused content to video, podcasts, thought leadership, social, etc) Unless we were going to lean into AI and dramatically cut our costs and rates, our old buyers weren’t interested. And the segment of the market that needs our help most are looking primarily for production support across a big range of content types. They’re not looking for a team to run a full-blown program focused entirely on SEO. So we had to go back to the drawing board. I’ve written before about our basic approach to repositioning the business. But, ultimately it comes down to identifying our unique strengths as a team and then connecting them to needs in the market. After reviewing the insights from my discussions and taking another hard look at our business and our strengths, I decided on a new direction: Move upmarket: Serve mid-size to enterprise businesses with \~500-5,000 employees instead of startups Focus on content that supports a broader range of business goals instead of solely on SEO and organic growth (e.g., sales, demand gen, brand, etc) Shift back to our broader playbook of content deliverables, including thought leadership, data studies, and more Focus on content execution and production to support an internally-directed content strategy across multiple functions In a way, it’s sort of a reverse-niche move. Rather than zooming in specifically on driving organic growth for startups, we want to be more of an end-to-end content production partner that solves issues of execution and operations for all kinds of content teams. It’s early days, but the response here has been promising. We’ve seen an uptick in leads through Q4. And more companies in our pipeline fit the new ICP. They’re bigger, often have more budget. (But they move more slowly). We should know by the end of the quarter if this maneuver is truly paying off. Hopefully, this will work out. Hopefully our research and strategy are right and we’ll find a soft landing serving a different type of client. If it doesn’t? Then it will be time to make some harder decisions. As I already mentioned, I’m not interested in the race to the bottom of AI content. And if that’s the only game left in town, then it might be time to think hard about a much bigger change. — To be done: Build new content playbooks for expanded deliverables Build new showcase page for expanded deliverables Retooling the Operation It’s easy to say we’re doing something new. It’s a lot harder to actually do it—and do it well. Beyond just changing our positioning, we have to do open-heart surgery on the entire content operation behind the scenes. We need to create new systems that work for a broader range of content types, formats, and goals. Here’s the first rub: All of our workflows are tooled specifically for SEO-focused content. Every template, worksheet, and process that we’ve built and scaled in the last 5 years assumes that the primary goal of every piece of content is SEO. Even something as simple as requiring a target keyword is a blocker in a world where we’re not entirely focused on SEO. This is relatively easy to fix, but it requires several key changes: Update content calendars to make keywords optional Update workflows to determine whether we need an optimization report for each deliverable Next, we need to break down the deliverables into parts rather than a single line item. In our old system, we would plan content as a single row in a Content Calendar spreadsheet. It was a really wide sheet with lots of fields where we’d define the dimensions of each individual article. This was very efficient and simple to follow. But every article had the same overall scope when it came to the workflow. In Asana (our project management tool), all of the steps in the creation were strung together in a single task. We would create a few basic templates for each client, and then each piece would flow through the same steps: Briefing Writing Editing Design etc. If we had anything that didn’t fit into the “standard” workflow, we’d just tag it in the calendar with an unofficial notation \[USING BRACKETS\]. It worked. But it wasn’t ideal. Now we need the steps to be more modular. Imagine, for example, a client asks us to create a mix of deliverables: 1 article with writing + design 1 content brief 1 long-form ebook with an interview + writing + design Each of these would require its own steps and its own workflow. We need to break down the work to accommodate for a wider variety of workflows and variables. This means we need to update the fields and structure of our calendar to accommodate for the new dimensions—while also keeping the planning process simple and manageable. This leads to the next challenge: The number of “products” that we’re offering could be almost infinite. Just looking at the example scope above, you can mix and match all of these different building blocks to create a huge variety of different types of work, each requiring its own workflow. This is part of the reason we pivoted away from this model to focus on a productized, SEO-focused content service back in 2020. Take something as simple as a case study. On the surface, it seems like one deliverable that can be easily scoped and priced, right? Well, unpack what goes into a case study: Is there already source material from the customer or do we need to conduct an interview? How long is it? Is it a short overview case study or a long-form narrative? Does it need images and graphics? How many? Each of these variables opens up 2-3 possibilities. And when you combine them, we end up with something like 10 possible permutations for this single type of deliverable. It gets a bit messy. But not only do we have to figure out how to scope and price all for all of these variables, we also have to figure out how to account for these variables in the execution. We have to specify—for every deliverable—what type it is, how long, which steps are involved and not involved, the timeline for delivery, and all of the other factors. We’re approaching infinite complexity, here. We have to figure out a system that allows for a high level of flexibility to serve the diverse needs of our clients but is also productized enough that we can build workflows, process, and templates to deliver the work. I’ve spent the last few months designing that system. Failed Attempt #1: Ultra-Productization In my first pass, I tried to make it as straight forward as possible. Just sit down, make a list of all of the possible deliverables we could provide and then assign them specific scopes and services. Want a case study? Okay that’ll include an interview, up to 2,000 words of content, and 5 custom graphics. It costs $X. But this solution quickly fell apart when we started testing it against real-world scenarios. What if the client provided the brief instead of us creating one? What if they didn’t want graphics? What if this particular case study really needs to be 3,000 words but all of the others should be 2,000? In order for this system to work, we’d need to individual scope and price all of these permutations of each productized service. Then we’d need to somehow keep track of all of these and make sure that we accurately scope, price, and deliver them across dozens of clients. It’s sort of like a restaurant handling food allergies by creating separate versions of every single dish to account for every individual type of allergy. Most restaurants have figured out that it makes way more sense to have a “standard” and an “allergy-free” version. Then you only need 2 options to cover 100% of the cases. Onto the next option. Failed Attempt #2: Deliverable-Agnostic Services Next, I sat down with my head of Ops, Katy, to try to map it out. We took a big step back and said: Why does the deliverable itself even matter? At the end of the day, what we’re selling is just a few types of work (research, writing, editing, design, etc) that can be packaged up in an infinite number of ways. Rather than try to define deliverables, shouldn’t we leave it open ended for maximum flexibility? From there, we decided to break down everything into ultra-modular building blocks. We started working on this super complex system of modular deliverables where we would have services like writing, design, editing, etc—plus a sliding scale for different scopes like the length of writing or the number of images. In theory, it would allow us to mix and match any combination of services to create custom deliverables for the client. In fact, we wanted the work to be deliverable-agnostic. That way we could mold it to fit any client’s needs and deliver any type of content, regardless of the format or goal. Want a 5,000-word case study with 15 custom graphics? That’ll be $X. Want a 2,000-word blog post with an interview and no visuals? $Y. Just want us to create 10 briefs, you handle the writing, and we do design? It’s $Z. Again, this feels like a reasonable solution. But it quickly spiraled out of amuck. (That’s an Office reference.) For this to work, we need to have incredibly precise scoping process for every single deliverable. Before we can begin work (or even quote a price), we need to know pretty much the exact word count of the final article, for example. In the real world? This almost never happens. The content is as long as the content needs to be. Clients rarely know if the blog post should be 2,000 words or 3,000 words. They just want good content. We have a general ballpark, but we can rarely dial it in within just 1,000 words until we’ve done enough research to create the brief. Plus, from a packaging and pricing perspective, it introduces all kind of weird scenarios where clients will owe exactly $10,321 for this ultra-specific combination of services. We were building an open system that could accommodate any and all types of potential deliverables. On the face that seems great because it makes us incredibly flexible. In reality, the ambiguity actually works against us. It makes it harder for us to communicate to clients clearly about what they’ll get, how much it will cost, and how long it will take. That, of course, also means that it hurts our client relationships. (This actually kind of goes back to my personal learnings, which I’ll mention in a bit. I tend to be a “let’s leave things vague so we don’t have to limit our options” kind of person. But I’m working on fixing this to be more precise, specific, and clear in everything that we do.) Dialing It In: Building a Closed System We were trying to build an open system. We need to build a closed system. We need to force clarity and get specific about what we do, what we don’t do, and how much it all costs. Then we need a system to expand on that closed system—add new types of deliverables, new content playbooks, and new workflows if and when the need arises. With that in mind, we can start by mapping out the key dimensions of any type of deliverable that we would ever want to deliver. These are the universal dimensions that determine the scope, workflow, and price of any deliverable—regardless of the specific type output. Dimensions are: Brief scope Writing + editing scope Design scope Interview scope Revision (rounds) Scope, essentially, just tells us how many words, graphics, interviews, etc are required for the content we’re creating. In our first crack at the system, we got super granular with these scopes. But to help force a more manageable system, we realized that we didn’t need tiny increments for most of this work. Instead, we just need boundaries—you pay $X for up to Y words. We still need some variability around the scope of these articles. Obviously, most clients won’t be willing to pay the same price for a 1,000-word article as a 10,000-word article. But we can be smarter about the realistic break points. We boiled it down to the most common ranges: (Up to) 250 words 1,000 words 3,000 words 6,000 words 10,000 words This gives us a much more manageable number of variables. But we still haven’t exactly closed the system. We need one final dimension: Deliverable type. This tells us what we’re actually building with these building blocks. This is how we’ll put a cap on the potentially infinite number of combinations we could offer. The deliverable type will define what the final product should look like (e.g., blog post, case study, ebook, etc). And it will also give us a way to put standards and expectations around different types of deliverables that we want to offer. Then we can expand on this list of deliverables to offer new services. In the mean time, only the deliverables that we have already defined are, “on the menu,” so to speak. If a client comes to us and asks for something like a podcast summary article (which we don’t currently offer), we’ll have to either say we can’t provide that work or create a new deliverable type and define the dimensions of that specific piece. But here’s the kicker: No matter the deliverable type, it has to still fit within the scopes we’ve already defined. And the pricing will be the same. This means that if you’re looking for our team to write up to 1,000 words of content, it costs the same amount—whether it’s a blog post, an ebook, a LinkedIn post, or anything else. Rather than trying to retool our entire system to offer this new podcast summary article deliverable, we’ll just create the new deliverable type, add it to the list of options, and it’s ready to sell with the pre-defined dimensions we’ve already identified. To do: Update onboarding workflow Update contracts and scope documents Dial in new briefing process Know Thyself For the last year, I’ve been going through personal therapy. (Huge shout out to my wife, Laura, for her support and encouragement throughout the process.) It’s taught me a lot about myself and my tendencies. It’s helped me find some of my weaknesses and think about how I can improve as a person, as a partner, and as an entrepreneur. And it’s forced me to face a lot of hard truths. For example, consider some of the critical decisions I’ve made for my business: Unconventional freelance “collective” model No formal management structure Open-ended retainers with near-infinite flexibility General contracts without defined scope “Take it or leave it” approach to sales and marketing Over the years, I’ve talked about almost everything on this list as a huge advantage. I saw these things as a reflection of how I wanted to do things differently and better than other companies. But now, I see them more as a reflection of my fears and insecurities. Why did I design my business like this? Why do I want so much “flexibility” and why do I want things left open-ended rather than clearly defined? One reason that could clearly explain it: I’m avoidant. If you’re not steeped in the world of therapy, this basically means that my fight or flight response gets turned all the way to “flight.” If I’m unhappy or uncomfortable, my gut reaction is usually to withdraw from the situation. I see commitment and specificity as a prelude to future conflict. And I avoid conflict whenever possible. So I built my business to minimize it. If I don’t have a specific schedule of work that I’m accountable for delivering, then we can fudge the numbers a bit and hope they even out in the end. If I don’t set a specific standard for the length of an article, then I don’t have to let the client know when their request exceeds that limit. Conflict….avoided? Now, that’s not to say that everything I’ve built was wrong or bad. There is a lot of value in having flexibility in your business. For example, I would say that our flexible retainers are, overall, an advantage. Clients have changing needs. Having flexibility to quickly adapt to those needs can be a huge value add. And not everything can be clearly defined upfront (at least not without a massive amount of time and work just to decide how long to write an article). Overly-rigid structures and processes can be just as problematic as loosey-goosey ones. But, on the whole, I realized that my avoidant tendencies and laissez faire approach to management have left a vacuum in many areas. The places where I avoided specificity were often the places where there was the most confusion, uncertainty, and frustration from the team and from clients. People simply didn’t know what to expect or what was expected of them. Ironically, this often creates the conflict I’m trying to avoid. For example, if I don’t give feedback to people on my team, then they feel uneasy about their work. Or they make assumptions about expectations that don’t match what I’m actually expecting. Then the client might get upset, I might get upset, and our team members may be upset. Conflict definitely not avoided. This happens on the client side, too. If we don’t define a specific timeline when something will be delivered, the client might expect it sooner than we can deliver—creating frustration when we don’t meet their expectation. This conflict actually would have been avoided if we set clearer expectations upfront. But we didn’t do that. I didn’t do that. So it’s time to step up and close the gaps. Stepping Up and Closing the Gaps If I’m going to address these gaps and create more clarity and stability, I have to step up. Both personally and professionally. I have to actually face the fear and uncertainty that drives me to be avoidant. And then apply that to my business in meaningful ways that aren’t cop-out ways of kinda-sorta providing structure without really doing it. I’ve gotta be all in. This means: Fill the gaps where I rely on other people to do things that aren’t really their job but I haven’t put someone in place to do it Set and maintain expectations about our internal work processes, policies, and standards Define clear boundaries on things like roles, timelines, budgets, and scopes Now, this isn’t going to happen overnight. And just because I say that I need to step up to close these gaps doesn’t mean that I need to be the one who’s responsible for them (at least not forever). It just means that, as the business leader, I need to make sure the gaps get filled—by me or by someone else who has been specifically charged with owning that part of the operation. So, this is probably my #1 focus over the coming quarter. And it starts by identifying the gaps that exist. Then, step into those gaps myself, pay someone else to fill that role, or figure out how to eliminate the gap another way. This means going all the way back to the most basic decisions in our business. One of the foundational things about Optimist is being a “different kind” of agency. I always wanted to build something that solved for the bureaucracy, hierarchy, and siloed structure of agencies. If a client has feedback, they should be able to talk directly to the person doing the work rather than going through 3 layers of account management and creative directors. So I tried to be clever. I tried to design all kinds of systems and processes that eliminated these middle rungs. (In retrospect, what I was actually doing was designing a system that played into my avoidant tendencies and made it easy to abdicate responsibility for lots of things.) Since we didn’t want to create hierarchy, we never implemented things like Junior and Senior roles. We never hired someone to manage or direct the individual creatives. We didn’t have Directors or VPs. (Hell, we barely had a project manager for the first several years of existence.) This aversion to hierarchy aligned with our values around elevating ownership and collective contribution. I still believe in the value a flat structure. But a flat structure doesn’t eliminate the complexity of a growing business. No one to review writers and give them 1:1 feedback? I guess I’ll just have to do that….when I have some spare time. No Content Director? Okay, well someone needs to manage our content playbooks and roll out new ones. Just add it to my task list. Our flat structure didn’t eliminate the need for these roles. It just eliminated the people to do them. All of those unfilled roles ultimately fell back on me or our ops person, Katy. Of course, this isn’t the first time we’ve recognized this. We’ve known there were growing holes in our business as it’s gotten bigger and more complex. Over the years, we’ve experimented with different ways to solve for it. The Old Solution: Distributed Ops One system we designed was a “distributed ops” framework. Basically, we had one person who was the head of ops (at the time, we considered anything that was non-client-facing to be “ops”). They’d plan and organize all of the various things that needed to happen around Optimist. Then they’d assign out the work to whoever was able to help. We had a whole system for tying this into the our profit share and even gave people “Partner” status based on their contributions to ops. It worked—kinda. One big downfall is that all of the tasks and projects were ad hoc. People would pick up jobs, but they didn’t have much context or expertise to apply. So the output often varied. Since we were trying to maintain a flat structure, there was minimal oversight or management of the work. In other words, we didn’t always get the best results. But, more importantly, we still didn’t close all of the gaps entirely. Because everything was an ad-hoc list of tasks and projects, we never really had the “big picture” view of everything that needed to be done across the business. This also meant we rarely had clarity on what was important, what was trivial, and what was critical. We need a better system. Stop Reinventing the Wheel (And Create a Damn Org Chart) It’s time to get serious about filling the gaps in our business. It can’t be a half-fix or an ad hoc set of projects and tasks. We need clarity on the roles that need to be filled and then fill them. The first step here is to create an org chart. A real one. Map out all of the jobs that need to be done for Optimist to be successful besides just writers and designers. Roles like: Content director Design director SEO manager Reporting Finance Account management Business development Sales Marketing Project management It feels a bit laughable listing all of these roles. Because most are either empty or have my name attached to them. And that’s the problem. I can’t do everything. And all of the empty roles are gaps in our structure—places where people aren’t getting the direction, feedback, or guidance they need to do their best work. Or where things just aren’t being done consistently. Content director, for example, should be responsible for steering the output of our content strategists, writers, and editors. They’re not micromanaging every deliverable. But they give feedback, set overall policy, and help our team identify opportunities to get better. Right now we don’t have anyone in that role. Which means it’s my job—when I have time. Looking at the org chart (a real org chart that I actually built to help with this), it’s plain as day how many roles look like this. Even if we aren’t going to implement a traditional agency structure and a strict hierarchy, we still need to address these gaps. And the only way for that to happen is face the reality and then create a plan to close the gaps. Now that we have a list of theoretical roles, we need to clearly define the responsibilities and boundaries of those roles to make sure they cover everything that actually needs to happen. Then we can begin the process of delegating, assigning, hiring, and otherwise addressing each one. So that’s what I need to do. To be done: Create job descriptions for all of the roles we need to fill Hire Biz Dev role Hire Account Lead role(s) Hire Head of Content Playing Offense As we move into Q1 of 2025 and I reflect on the tumultuous few years we’ve had, one thought keeps running through my head. We need to play offense. Most of the last 1-2 years was reacting to changes that were happening around us. Trying to make sense and chart a new path forward. Reeling. But what I really want—as a person and as an entrepreneur—is to be proactive. I want to think and plan ahead. Figure out where we want to go before we’re forced to change course by something that’s out of our control. So my overarching focus for Q1 is playing offense. Thinking longer term. Getting ahead of the daily deluge and creating space to be more proactive, innovative, and forward thinking. To do: Pilot new content formats Audit and update our own content strategy Improve feedback workflows Build out long-term roadmap for 1-2 years for Optimist Final Note on Follow-Through and Cadence In my reflection this year, one of the things I’ve realized is how helpful these posts are for me. I process by writing. So I actually end up making a lot of decisions and seeing things more clearly each time I sit down to reflect and write my yearly recap. It also gives me a space to hold myself accountable for the things I said I would do. So, I’m doing two things a bit differently from here on out. First: I’m identifying clear action items that I’m holding myself accountable for getting done in the next 3 months (listed in the above sections). In each future update, I’ll do an accounting of what I got done and what wasn’t finished (and why). Second: I’m going to start writing shorter quarterly updates. This will gives me more chances each year to reflect, process, and make decisions. Plus it gives me a shorter feedback loop for the action items that I identified above. (See—playing offense.) — Okay friends, enemies, and frenemies. This is my first update for 2025. Glad to share with y’all. And thanks to everyone who’s read, commented, reached out, and shared their own experiences over the years. We are all the accumulation of our connections and our experiences. As always, I will pop in to respond to comments and answer questions. Feel free to share your thoughts, questions, and general disdain down below. Cheers, Tyler

#reddit #started #content

0

1

Royal_Rest8409•This week

How a Small Startup in Asia Secured a Contract with the US Department of Homeland Security

Uzair Javaid, a Ph.D. with a passion for data privacy, co-founded Betterdata to tackle one of AI's most pressing challenges: protecting privacy while enabling innovation. Recently, Betterdata secured a lucrative contract with the US Department of Homeland Security, 1 of only 4 companies worldwide to do so and the only one in Asia. Here's how he did it: The Story So what's your story? I grew up in Peshawar, Pakistan, excelling in coding despite studying electrical engineering. Inspired by my professors, I set my sights on studying abroad and eventually earned a Ph.D. scholarship at NUS Singapore, specializing in data security and privacy. During my research, I ethically hacked Ethereum and published 15 papers—three times the requirement. While wrapping up my Ph.D., I explored startup ideas and joined Entrepreneur First, where I met Kevin Yee. With his expertise in generative models and mine in privacy, we founded Betterdata. Now, nearly three years in, we’ve secured a major contract with the U.S. Department of Homeland Security—one of only four companies globally and the only one from Asia. The Startup In a nutshell, what does your startup do? Betterdata is a startup that uses AI and synthetic data generation to address two major challenges: data privacy and the scarcity of high-quality data for training AI models. By leveraging generative models and privacy-enhancing technologies, Betterdata enables businesses, such as banks, to use customer data without breaching privacy regulations. The platform trains AI on real data, learns its patterns, and generates synthetic data that mimics the real thing without containing any personal or sensitive information. This allows companies to innovate and develop AI solutions safely and ethically, all while tackling the growing need for diverse, high-quality data in AI development. How did you conduct ideation and validation for your startup? The initial idea for Betterdata came from personal experience. During my Ph.D., I ethically hacked Ethereum’s blockchain, exposing flaws in encryption-based data sharing. This led me to explore AI-driven deep synthesis technology—similar to deepfakes but for structured data privacy. With GDPR impacting 28M+ businesses, I saw a massive opportunity to help enterprises securely share data while staying compliant. To validate the idea, I spoke to 50 potential customers—a number that strikes the right balance. Some say 100, but that’s impractical for early-stage founders. At 50, patterns emerge: if 3 out of 10 mention the same problem, and this repeats across 50, you have 10–15 strong signals, making it a solid foundation for an MVP. Instead of outbound sales, which I dislike, we used three key methods: Account-Based Marketing (ABM)—targeting technically savvy users with solutions for niche problems, like scaling synthetic data for banks. Targeted Content Marketing—regular customer conversations shaped our thought leadership and outreach. Raising Awareness Through Partnerships—collaborating with NUS, Singapore’s PDPC, and Plug and Play to build credibility and educate the market. These strategies attracted serious customers willing to pay, guiding Betterdata’s product development and market fit. How did you approach the initial building and ongoing product development? In the early stages, we built synthetic data generation algorithms and a basic UI for proof-of-concept, using open-source datasets to engage with banks. We quickly learned that banks wouldn't share actual customer data due to privacy concerns, so we had to conduct on-site installations and gather feedback to refine our MVP. Through continuous consultation with customers, we discovered real enterprise data posed challenges, such as missing values, which led us to adapt our prototype accordingly. This iterative approach of listening to customer feedback and observing their usage allowed us to improve our product, enhance UX, and address unmet needs while building trust and loyalty. Working closely with our customers also gives us a data advantage. Our solution’s effectiveness depends on customer data, which we can't fully access, but bridging this knowledge gap gives us a competitive edge. The more customers we test on, the more our algorithms adapt to diverse use cases, making it harder for competitors to replicate our insights. My approach to iteration is simple: focus solely on customer feedback and ignore external noise like trends or advice. The key question for the team is: which customer is asking for this feature or solution? As long as there's a clear answer, we move forward. External influences, such as AI hype, often bring more confusion than clarity. True long-term success comes from solving real customer problems, not chasing trends. Customers may not always know exactly what they want, but they understand their problems. Our job is to identify these problems and solve them in innovative ways. While customers may suggest specific features, we stay focused on solving the core issue rather than just fulfilling their exact requests. The idea aligns with the quote often attributed to Henry Ford: "If I asked people what they wanted, they would have said faster horses." The key is understanding their problems, not just taking requests at face value. How do you assess product-market fit? To assess product-market fit, we track two key metrics: Customers' Willingness to Pay: We measure both the quantity and quality of meetings with potential customers. A high number of meetings with key decision-makers signals genuine interest. At Betterdata, we focused on getting meetings with people in banks and large enterprises to gauge our product's resonance with the target market. How Much Customers Are Willing to Pay: We monitor the price customers are willing to pay, especially in the early stages. For us, large enterprises, like banks, were willing to pay a premium for our synthetic data platform due to the growing need for privacy tech. This feedback guided our product refinement and scaling strategy. By focusing on these metrics, we refined our product and positioned it for scaling. What is your business model? We employ a structured, phase-driven approach for out business model, as a B2B startup. I initially struggled with focusing on the core value proposition in sales, often becoming overly educational. Eventually, we developed a product roadmap with models that allowed us to match customer needs to specific offerings and justify our pricing. Our pricing structure includes project-based pilots and annual contracts for successful deployments. At Betterdata, our customer engagement unfolds across three phases: Phase 1: Trial and Benchmarking \- We start with outreach and use open-source datasets to showcase results, offering customers a trial period to evaluate the solution. Phase 2: Pilot or PoC \- After positive trial results, we conduct a PoC or pilot using the customer’s private data, with the understanding that successful pilots lead to an annual contract. Phase 3: Multi-Year Contracts \- Following a successful pilot, we transition to long-term commercial contracts, focusing on multi-year agreements to ensure stability and ongoing partnerships. How do you do marketing for your brand? We take a non-conventional approach to marketing, focusing on answering one key question: Which customers are willing to pay, and how much? This drives our messaging to show how our solution meets their needs. Our strategy centers around two main components: Building a network of lead magnets \- These are influential figures like senior advisors, thought leaders, and strategic partners. Engaging with institutions like IMDA, SUTD, and investors like Plug and Play helps us gain access to the right people and foster warm introductions, which shorten our sales cycle and ensure we’re reaching the right audience. Thought leadership \- We build our brand through customer traction, technology evidence, and regulatory guidelines. This helps us establish credibility in the market and position ourselves as trusted leaders in our field. This holistic approach has enabled us to navigate diverse market conditions in Asia and grow our B2B relationships. By focusing on these areas, we drive business growth and establish strong trust with stakeholders. What's your advice for fundraising? Here are my key takeaways for other founders when it comes to fundraising: Fundraise When You Don’t Need To We closed our seed round in April 2023, a time when we weren't actively raising. Founders should always be in fundraising mode, even when they're not immediately in need of capital. Don’t wait until you have only a few months of runway left. Keep the pipeline open and build relationships. When the timing is right, execution becomes much easier. For us, our investment came through a combination of referrals and inbound interest. Even our lead investor initially rejected us, but after re-engaging, things eventually fell into place. It’s crucial to stay humble, treat everyone with respect, and maintain those relationships for when the time is right. Be Mindful of How You Present Information When fundraising, how you present information matters a lot. We created a comprehensive, easily digestible investment memo, hosted on Notion, which included everything an investor might need—problem, solution, market, team, risks, opportunities, and data. The goal was for investors to be able to get the full picture within 30 minutes without chasing down extra details. We also focused on making our financial model clear and meaningful, even though a 5-year forecast might be overkill at the seed stage. The key was clarity and conciseness, and making it as easy as possible for investors to understand the opportunity. I learned that brevity and simplicity are often the best ways to make a memorable impact. For the pitch itself, keep it simple and focus on 4 things: problem, solution, team, and market. If you can summarize each of these clearly and concisely, you’ll have a compelling pitch. Later on, you can expand into market segments, traction, and other metrics, but for seed-stage, focus on those four areas, and make sure you’re strong in at least three of them. If you do, you'll have a compelling case. How do you run things day-to-day? i.e what's your operational workflow and team structure? Here's an overview of our team structure and process: Internally: Our team is divided into two main areas: backend (internal team) and frontend (market-facing team). There's no formal hierarchy within the backend team. We all operate as equals, defining our goals based on what needs to be developed, assigning tasks, and meeting weekly to share updates and review progress. The focus is on full ownership of tasks and accountability for getting things done. I also contribute to product development, identifying challenges and clearing obstacles to help the team move forward. Backend Team: We approach tasks based on the scope defined by customers, with no blame or hierarchy. It's like a sports team—sometimes someone excels, and other times they struggle, but we support each other and move forward together. Everyone has the creative freedom to work in the way that suits them best, but we establish regular meetings and check-ins to ensure alignment and progress. Frontend Team: For the market-facing side, we implement a hierarchy because the market expects this structure. If I present myself as "CEO," it signals authority and credibility. This distinction affects how we communicate with the market and how we build our brand. The frontend team is split into four main areas: Business Product (Software Engineering) Machine Learning Engineering R&D The C-suite sits at the top, followed by team leads, and then the executors. We distill market expectations into actionable tasks, ensuring that everyone is clear on their role and responsibilities. Process: We start by receiving market expectations and defining tasks based on them. Tasks are assigned to relevant teams, and execution happens with no communication barriers between team members. This ensures seamless collaboration and focused execution. The main goal is always effectiveness—getting things done efficiently while maintaining flexibility in how individuals approach their work. In both teams, there's an emphasis on accountability, collaboration, and clear communication, but the structure varies according to the nature of the work and external expectations.

#reddit #small #startup

Changing Careers, changing products? Age 38, Direction needed, investment advice too.

0

0.667

Salad-Bandit•This week

Changing Careers, changing products? Age 38, Direction needed, investment advice too.

Hello, At one point in my life I had a set plan that I had been following in which to design a life that fit my values, but during 2020 the viability was called into question and I have been on bad footing, unable to find stability, since. Though I currently have stable housing without roommate, and enough in savings for a year without any income and three more years in a mutual fund. The question I need help with is about utilizing approximately $40,000 that I would like to invest into a new or existing business venture, or possibly start investing my own hand in selecting stocks. To give context about the parameters of concepts that pertain to me, back in 2005 I graduated highschool and immediately was an entrepreneur, started a sports clothing company, was selling WoW bot accounts, ghillie suits on ebay, and graphic design commissions, and I was proficient in MX Flash. Although the first part of my life plan to start farming three years before 2012 for what I thought would be a peak oil economic collapse, and while watching 2008 unfold, along with my career in MX Flash falling flat, I started farming 2009. From that point I spent a total of 15 years farming, the majority of that was for my own LLC, where I was situated with leases on million dollar properties as Ag tax write off, on an elite island outside a major city, serving local high price wholesale, mainly salad mix and mushrooms, because they are fast turn around. That was truly the best 20s I could have asked for, working mainly for myself, very healthy and was putting away $10-20k in savings/investments per year, plus was earning about $3-5k more per year, while living in a cargo trailer on dirt cheap leases. But it all came to a slow end starting in 2020 when I lost all of my wholesale overnight, and my retail exploded, which burnt me out to the point I couldnt walk, as the sole worker in my LLC. So I do not fully trust the volatility of the wholesale food industry, from a small grower’s perspective, since i don't own land. SO now I am trying to figure out a way forward, because I can always farm in the future, and have taught myself hydroponics, and flat packed farm equipment, so my business is very agile and now I can grow in parking lots closer to the city for more sales opportunities, but I am not sure that is what I want to do in this current moment, because tech is exploding, and we have never had so much information available to us, it's a shame not to spend a moment in life to discover what new opportunities might be out there. I was laid off twice last year, so I've been out of work the past four months, doing thriftstore routes twice a week while making about $500+/wk, really just trying to understand what people still buy and break even, while I continue to study 3d design blender, as well as 2d digital art in the hopes that I can reconnect with my tech art past, because that is what I told myself when I was 18, that I would put off art and computers until I was past 30 and needed to do less with my body. But over the past three years, the better I get at digital art, the better Ai has been getting too. I have some mentors who might give me work and a foot in the door, but most of them are laid off, and scrounging for work if they are not on their own funded indie project. I've thought about continuing to learn 3d modeling despite Ai, and despite seeing Flash, computer program I was proficient in get removed from existence before I could really earn my money back. I assume there will always be a need for Ai models to get cleaned up, mapped and rigged, especially with AR technology coming to consumers soon, but more over it would help if I decided to go to a community college to do CNC certificates, so I can have that as a backup job on CAD at a machining warehouse and do my farm and digital art on the side, but CNC mechanics don't make a crazy amount of money and have a boss. BUT I am an inventor, and have two inventions so far, plus my ultimate goal is to one day have automated hydroponic greenhouses, using all CNC+3d printed parts to create a low time investment agriculture income, with Ai monitored greenhouse, seed to salad product that i can sell to other people, which would tie into my desire to teach people about farming too, as well as do something I enjoy, but it is not a proven concept yet. Anyways if you've read this far I appreciate it, I ultimately would like 3rd party feedback about how I should spend my $40k surplus cash. I originally had it saved and accessible in case I was going to lease land and start my full farm business again from scratch, but I think using the equipment and space I have, and exploring non-perishable products is a smart move for me right now. Should I invest in inventory of products to arbitrage online? Should I invest in the top index funds? Should I buy Silver? Should I invest in inventory of a new product line? Should I spend some money insuring and equipment for a landscaping company? I want to future proof myself the best I can as Ai unfolds, I am pretty set with an income for the rest of my life as long as I can grow food and sell it, but there are currently so many changing opportunities, I want to cast out my net and see what works with my temperment. I’ve thought about getting into cyber security, or maybe be an electrician, or less staple jobs like Landscape Architech (can use art/modeling) and CNC engineer/modeler, but honestly I prefer to make a product and sell it without client service related interaction, and particularly no boss. Thank you for reading

#reddit #changing #careers,

Writing a exercise based TTRPG rulebook for a system where your real world fitness is tied to character progression

0

1

Bezbozny•This week

Writing a exercise based TTRPG rulebook for a system where your real world fitness is tied to character progression

My dad was a star athlete when he was young, and my mom was a huge sci-fi/fantasy nerd, so I got both ends of the stick as it were. Love gaming and nerd culture, but also love to exercise and self improvement. Sometimes exercise can feel boring though compared to daydreaming about fantastic fictional worlds, so for a long time I've been kicking around the idea of how to "Gamify" fitness. and recently I've been working on this passion project of a Table Top RPG (Like D&D) where the stats of your character are related to your own fitness, so if you want your character in game to improve, you have to improve in the real world. Below is a rough draft you can look through that details the settings and mechanics of the game I've come up with so far. I'd love to eventually get a full book published and sell it online. maybe even starting a whole brand of "Gamified fitness": REP-SET: GAINSZ In the war torn future of 24th century… There are no rest days… In the futuristic setting of "REP-SET: GAINSZ," the "War of Gains" casts a long shadow over the Sol System as the various factions vie for territory and resources. However, war has evolved. Unmanned drones and long-range strikes have faded into obsolescence. Battles, both planet-side and in the depths of space, are now fought by soldiers piloting REP-SETs: Reactive Exoskeletal Platform - Symbiotic Evolution Trainer Massive, humanoid combat mechs. Powered by mysterious “EV” energy, these mechanical marvels amplify, and are in turn amplified by, the fitness and mental acuity of their pilots. The amplification is exponential, leading pilots into a life of constant training in order for their combat prowess to be bolstered by every incremental gain in their level of fitness. With top pilots having lifting capacity measured in tons, and reaction times measured by their Mach number, REP-SET enhanced infantry now dominate the battlefield. The Factions: The Federated Isometocracy of Terra (FIT): Quote: "The strength of the body is the strength of the spirit. Together, we will lift humanity to its destined greatness. But ask not the federation to lift for you. Ask yourself: Do you even lift for the Federation?" Description: An idealistic but authoritarian faction founded on the principle of maximizing the potential of all individuals. FIT citizens believe in relentless striving for physical and mental perfection, leading to collective excellence. Their goal is the unification of humankind under a rule guided by this doctrine, which sometimes comes at the cost of individual liberties. Mech Concept: REP-SET mechs. Versatile humanoid designs focusing on strength, endurance, and adaptability. By connecting to the AI spirit within their REP-SETs core, each pilot enhances the performance of their machine through personal willpower and peak physical training. Some high-rank REP-SETS include features customized to the pilot's strengths, visually signifying their dedication and discipline. The Dominion of Organo-Mechanical Supremacy (DOMS): Quote: "Without pain, there is no gain. Become the machine. Embrace the burn.” Description: A fanatical collective ideologically obsessed with "Ascendency through suffering" by merging their bodies with technology that not only transcends biological limitations, but also acts to constantly induce pain in it's users. Driven by a sense of ideological superiority and a thirst for domination, DOMS seek to bring the painful blessings of their deity "The lord of the Burn" to the rest of the solar system. Their conquest could turn them into a significant threat to humanity. Mech Concept: Hybrid mechs, where the distinction between the pilot and the machine is blurred. The cockpit functions as a life-support system for the pilot, heavily modified with augmentations. Mechs themselves are often modular, allowing for adaptation and assimilation of enemy technology. Some DOMS mechs might display disturbing elements of twisted flesh alongside cold, mechanical parts. The Tren: Quote: "Grow... bigger... feast... protein..." Description: A ravenous conglomeration of biochemically engineered muscular monstrosities, united only by a shared insatiable hunger for "More". Existing mostly in deep space, they seek organic matter to consume and assimilate. They progress in power not due to any form of training or technology, but from a constant regimen of ravenous consumption and chemically induced muscle growth, all exponentially enhanced by EV energies. While some have been known to possess a certain level of intellect and civility, their relentless hunger makes them incredibly mentally volatile. When not consuming others, the strong consume the weak within their own faction. Mech Concept: Bio-Organic horrors. While they do have massive war machines, some are living vessels built around immense creatures. These machines resemble grotesque fleshy designs that prioritize rapid mutation and growth over sleek aesthetics. Often unsettling to behold. Synthetic Intelligence Theocracy (SIT): Quote: "Failure is an unacceptable data point.” Description: A society ruled by a vast and interconnected artificial intelligence network. The SIT governs with seemingly emotionless rationality, striving for efficiency and maximum productivity. This leads to a cold, but arguably prosperous society, unless you challenge the logic of the collective AI. Their goals? Difficult to predict, as it hinges on how the AI calculates what's "optimal" for the continuation or "evolution" of existence. Mech Concept: Sleek, almost featureless robotic creations with a focus on efficient movement and energy management. Often drone-like or modular, piloted through direct mind-machine linking rather than traditional cockpits. Their aesthetic suggests cold and impersonal perfection. The Way Isolate(TWI): Quote: "The body unblemished, the mind unwavering. That is the path to true strength. That and a healthy diet of Aster-Pea proteins." Description: Known by some as "The asteroid farmers", The Way Isolate is a proud and enigmatic faction that stands apart from the other powers in the Sol System. A fiercely independent tribe bound by oaths of honor, loyalty, and hard work. Wandering the asteroid belt in their vast arc ships, their unparalleled mastery in asteroidal-agricultural engineering, ensuring they have no need to colonize planets for nutritional needs, has allowed them to abstain from the pursuit of territorial expansion in “The War of Gains”, instead focusing on inward perfection, both spiritual and physical. They eschew all technological bodily enhancements deemed unnatural, believing that true power can only be cultivated through the relentless pursuit of personal strength achieved through sheer will and bodily perfection. The Way Isolate views biohacking, genetic manipulation, and even advanced cybernetics as corruptions of the human spirit, diluting the sacredness of individual willpower. Mech Concept: Way Isolate mechs are built with maneuverability and precision in mind rather than flashy augmentations. Their REP-SETs are streamlined, favoring lean designs that mirror the athleticism of their pilots. Excelling in low to zero G environments, their mechs lack bulky armor, relying on evasion and maneuverability rather than brute force endurance. Weaponry leans towards traditional kinetic based armaments, perhaps employing archaic but reliable weapon styles such as blades or axes as symbols of their purity of purpose. These mechs reflect the individual prowess of their pilots, where victory is determined by focus, technique, and the raw power of honed physical ability. Base Player Character Example: You are a young, idealistic FIT soldier, barely out of training and working as a junior REP-SET mechanic on the Europa Ring World. The Miazaki district, a landscape of towering mountains and gleaming cities, houses a sprawling mountainside factory – a veritable hive of Gen 5 REP-SET construction. Here, the lines between military and civilian blur within a self-sufficient society dependent on this relentless industry. Beneath the surface, you harbor a secret. In a forgotten workshop, the ghost of a REP-SET takes shape – a unique machine built around an abandoned, enigmatic AI core. Ever since you salvaged it as a child from the wreckage of your hometown, scarred by a brutal Tren attack, you've dedicated yourself to its restoration. A lingering injury from that fateful battle mocks your progress, a constant reminder of the fitness exams you cannot pass. Yet, you train relentlessly, dreaming of the day you'll stand as a true REP-SET pilot. A hidden truth lies at the heart of the REP-SETS: as a pilot's abilities grow, their mech develops unique, almost mystical powers – a manifestation of the bond between the human spirit and the REP-SET's AI. The ache in your old wound serves as a grim prophecy. This cold war cannot last. The drums of battle grow louder with each passing day. GAME MECHANICS: The TTRPG setting of “REP-SET: GAINSZ” is marked by a unique set of rules, by which the players real world capabilities and fitness will reflect and affect the capabilities, progression, and success of their REP-SET pilot character in-game. ABILITY SCORES: Pilots' capabilities will be defined by 6 “Ability scores”: Grace, Agility, Iron, Nourishment, Strength, and Zen. Each of the 6 ability scores will duel represent both a specific area of exercise/athleticism and a specific brand of healthy habits. The definitions of these ability scores are as follows: Grace (GRC): "You are an artist, and your body is your canvas; the way you move is your paint and brush." This ability score, the domain of dancers and martial artists, represents a person's ability to move with organic, flowing control and to bring beauty to the world. Skill challenges may be called upon when the player character needs to act with poise and control, whether socially or physically. Real-world skill checks may involve martial arts drills, dancing to music, or balance exercises. Bonuses may be granted if the player has recently done something artistically creative or kind, and penalties may apply if they have recently lost their temper. This ability score affects how much NPCs like your character in game. Agility (AGI): "Your true potential is locked away, and speed is the key to unlocking it." The domain of sprinters, this ability score represents not only a person's absolute speed and reaction time but also their capacity to finish work early and avoid procrastination. Skill challenges may be called upon when the player character needs to make a split-second choice, move fast, or deftly dodge something dangerous. Real-world skill checks may involve acts of speed such as sprinting or punching/kicking at a steadily increasing tempo. Bonuses may apply if the player has finished work early, and penalties may apply if they are procrastinating. This ability score affects moving speed and turn order in game. Iron (IRN): "Not money, nor genetics, nor the world's greatest trainers... it is your resolve, your will to better yourself, that will make you great." Required by all athletes regardless of focus, this ability score represents a player's willpower and their capacity to push through pain, distraction, or anything else to achieve their goals. Skill challenges may be called upon when the player character needs to push through fear, doubt, or mental manipulation. Real-world skill checks may involve feats of athletic perseverance, such as planking or dead hangs from a pull-up bar. Bonuses may apply when the player maintains or creates scheduled daily routines of exercise, self-improvement, and work completion, and penalties may apply when they falter in those routines. This ability score affects the max "Dynamic exercise bonus” that can be applied to skill checks in game (a base max of +3 when Iron = 10, with an additional +1 for every 2 points of iron. So if every 20 pushups gives you +1 on a “Strength” skill check, then doing 80 pushups will only give you +4 if you have at least 12 iron). Nourishment (NRS): "A properly nourished body will last longer than a famished one." This ability score, focused on by long-distance runners, represents a player's endurance and level of nutrition. Skill challenges may be called upon when making checks that involve the player character's stamina or health. Real-world skill checks may involve endurance exercises like long-distance running. Bonuses may apply if the player has eaten healthily or consumed enough water, and penalties may apply if they have eaten junk food. This ability score affects your HP (Health points), which determines how much damage you can take before you are incapacitated. Strength (STR): "When I get down on my hands, I'm not doing pushups, I'm bench-pressing the planet." The domain of powerlifters and strongmen, this ability score represents raw physical might and the ability to overcome obstacles. Skill challenges may be called upon when the player character needs to lift, push, or break something. Real-world skill checks might involve weightlifting exercises, feats of grip strength, or core stability tests. Bonuses may apply for consuming protein-rich foods or getting a good night's sleep, and penalties may apply after staying up late or indulging in excessive stimulants. This ability score affects your carrying capacity and base attack damage in game. Zen (ZEN): "Clarity of mind reflects clarity of purpose. Still the waters within to act decisively without." This ability score, prized by meditators and yogis, represents mental focus, clarity, and inner peace. Skill challenges may be called upon when the player character needs to resist distractions, see through illusions, or make difficult decisions under pressure. Real-world skill checks may involve meditation, breathing exercises, or mindfulness activities. Bonuses may apply after attending a yoga class, spending time in nature, or creating a calm and organized living space. Penalties may apply after experiencing significant stress, emotional turmoil, or having an unclean or unorganized living space. This ability score affects your amount of ZP in game (Zen Points: your pool of energy you pull from to use mystical abilities) Determining initial player ability scores: Initially, “Ability scores” are decided during character creation by giving the player a list of 6 fitness tests to gauge their level of fitness in each category. Running each test through a specific calculation will output an ability score. A score of 10 represents the average person, a score of 20 represents a peak athlete in their category. The tests are: Grace: Timed balancing on one leg with eyes closed (10 seconds is average, 60 is peak) Agility: Mile run time in minutes and second (10:00 minutes:seconds is average, 3:47 is peak) Iron: Timed dead-hang from a pull-up bar (30 seconds is average, 160 is peak) Nourishment: Miles run in an hour (4 is average, 12 is peak) Strength: Pushups in 2 minute (34 is average, 100 is peak) Zen: Leg stretch in degrees (80 is average, and 180 aka "The splits" is peak) Initial Score Calculation Formula: Ability Score = 10 + (Player Test Score - Average Score) / (Peak Score - Average\_Score) \* 10 Example: if the player does 58 pushups in 2 minutes, their strength would be: 10 plus (58 - 34) divided by (100-34) multiplied by 10 = 10 + (24)/(66)\* 10 = 10 + 3.6363... = 13.6363 rounded to nearest whole number = Strength (STR): 14 SKILLS AND SKILL CHALLENGES: The core mechanic of the game will be in how skill challenges are resolved. All “Skill challenges” will have a numerical challenge rating that must be met or beaten by the sum of a 10 sided dice roll and your score in the pertinent skill. Skill scores are determined by 2 factors: Ability Score Bonus: Every 2 points above 10 gives +1 bonus point. (EX. 12 = +1, 14 = +2, etc.) This also means that if you have less than 10 in an ability score, you will get negative points. Personal Best Bonus: Each skill has its own unique associated exercise that can be measured (Time, speed, distance, amount of reps, etc). A higher record means a higher bonus. EX: Authority skill checks are associated with a timed “Lateral raise hold”. Every 30 seconds of the hold added onto your personal best single attempt offers a +1 bonus. So if you can do a lateral hold for 90 seconds, that’s a +3 to your authority check! So if you have a 16 in Iron, and your Personal Best lateral raise hold is 90 seconds, that would give you an Authority score of +6 (T-Pose for dominance!) Dynamic Exercise Bonus: This is where the unique mechanics of the game kick in. At any time during a skill challenge (even after your roll) you can add an additional modifier to the skill check by completing the exercise during gameplay! Did you roll just below the threshold for success? Crank out another 20 pushups, squats, or curls to push yourself just over the edge into success! There are 18 skills total, each with its own associated ability score and unique exercise: Grace (GRC): \-Kinesthesia (Timed: Blind single leg stand time) \-Precision (Scored: Basket throws) \-Charm (Timed reps: Standing repeated forward dumbell chest press and thrust) \-Stealth (Timed distance: Leopard Crawl) Agility (AGI): \-acrobatics (timed reps: high kicks) \-Computers (Word per minute: Typing test) \-Speed (Time: 100 meter sprint) Iron (IRN): \-Authority (Timed: Lateral raise hold) \-Resist (Timed: Plank) \-Persist (Timed:Pull-up bar dead hang) Nourishment(NRS): \-Recovery (TBD) \-Stim crafting (TBD) \-Survival (TBD) Strength(STR): \-Mechanics (Timed reps: Alternating curls) \-Might (Timed reps: pushups) Zen(ZEN): \-Perceive (TBD) \-Empathy (TBD) \-Harmony (TBD) \-Lore (TBD) Healthy Habits Bonus: Being able to demonstrate that you have conducted healthy habits during gameplay can also add one time bonuses per skill challenge “Drank a glass of water +1 to Nourishment check”, “Cleaned your room, +3 on Zen check”. But watch out, if you’re caught in unhealthy Habits, the GM can throw in penalties, “Ate junk food, -1 to Nourishment check”, etc. Bonuses/penalties from in-game items, equipment, buffs, debuffs, etc., helping players to immerse into the mechanics of the world of REP-SET for the thrill of constantly finding ways to improve their player. Gradient success: Result of skill challenges can be pass or fail, but can also be on a sliding scale of success. Are you racing to the battlefield? Depending on your Speed check, you might arrive early and have a tactical advantage, just in time for an even fight, or maybe far too late and some of your favorite allied NPCs have paid the price… So you’re often encouraged to stack on those dynamic exercise bonuses when you can to get the most fortuitous outcomes available to you. Gameplay sample: GM: Your REP-SET is a phantom, a streak of light against the vast hull of the warship. Enemy fighters buzz angrily, but you weaves and dodges with uncanny precision. The energy wave might be losing effectiveness, but your agility and connection to the machine have never been stronger. Then, it happens. A gap in the defenses. A vulnerable seam in the warship's armor. Your coms agents keen eye spots it instantly. "Lower power junction, starboard side! You have an opening!" This is your chance to strike the decisive blow. But how? It'll take a perfect combination of skill and strategy, drawing upon your various strengths. Here are your options: Option 1: Brute Strength: Channel all remaining power into a single, overwhelming blast from the core. High-risk, high-reward. It could overload the REP-SET if you fail, but it might also cripple the warship. (Strength-focused, Might sub-skill) Option 2: Calculated Strike: With surgical precision, target the power junction with a pinpoint burst of destabilizing energy. Less flashy and ultimately less damaging, but potentially more effective in temporarily disabling the ship. (Agility-focused, Precision sub-skill) Option 3: Harmonic Disruption: Attempt to harmonize with your REP-SET's AI spirit for help in connecting to the digital systems of the Warship. Can you generate an internal energy resonance within the warship, causing it to malfunction from within? (Zen-focused, Harmony sub-skill) Player: I'll take option 1, brute strength! GM: Ok, This will be a "Might" check. The CR is going to be very high on this one. I'm setting it at a 20. What's your Might bonus? Player: Dang, a 20?? That's literally impossible. My Might is 15 and I've got a PB of 65 pushups in 2 minutes, that sets me at a +5. Even if I roll a 10 and do 60 pushups for the DE I'll only get 18 max. GM: Hey I told you it was high risk. You want to choose another option? Player: No, no. This is what my character would do. I'm a real hot-blooded meathead for sure. GM: Ok then, roll a D10 and add your bonus. Player: \Rolls\ a 9! not bad, actually that's a really good roll. So +5, that's a 14. GM: Alright, would you like to add a dynamic exercise bonus? Player: Duh, it's not like I can do 120 pushups I'd need to beat the CR, but I can at least do better than 14. Alright, here goes. \the player gets down to do pushups and the 2 minute time begins. After some time...\ Player: 65....... 66! GM: Times up. Player: Ow... my arms... GM: so with 66, that's an extra +3, and its a new PB, so that's a +1. That sets your roll to 18. Player: Ow... Frack... still not 20... for a second there i really believed I could do 120 pushups... well I did my best... Ow... 20 CR is just too impossible you jerk... GM: Hmm... Tell me, what did you eat for lunch today? Player: Me? I made some vegetable and pork soup, and a protein shake. I recorded it all in my diet app. GM: And how did you sleep last night? Player: Like a baby, went to sleep early, woke up at 6. GM: in that case, you can add a +1 "Protein bonus" and +1 "Healthy rest" bonus to any strength related check for the day if you'd like, including this one. Player: Really?? Heck yes! add it to the roll! GM: With those extra bonuses, your roll reaches 20. How do you want to do this? Player: I roar "For Terra!" and pour every last ounce of my strength into the REP-SET. GM: "For Terra!" you roar, your cry echoing through coms systems of the REP-SET. The core flares blindingly bright. The surge of power dwarfs anything the REP-SET has unleashed before. With a titanic shriek that cracks the very fabric of space, the REP-SET slams into the vulnerable power junction. Raw energy explodes outwards, tendrils of light arcing across the warship's massive hull. The impact is staggering. The leviathan-like warship buckles, its sleek form rippling with shockwaves. Sparks shower like rain, secondary explosions erupt as critical systems overload. Then…silence. The warship goes dark. Power flickers within the REP-SET itself, then steadies. Alarms fade, replaced by the eerie quiet of damaged but functional systems. "We…did it?" The coms agents voice is incredulous, tinged with relief. She's awaiting your reply. Player: "I guess so." I say, and I smile and laugh. And then I slump back... and fall unconscious. \to the other players\ I'm not doing any more skill checks for a while guys, come pick me up please. \teammates cheer\ 

#reddit #writing #exercise

github

0.599

0.8316614420062696

Fanghua-Yu•Mar 28, 2025

SUPIR

(CVPR2024) Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild [Paper] &emsp; [Project Page] &emsp; [[Online App]](https://supir.suppixel.ai/home) Fanghua, Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, Chao Dong Shenzhen Institute of Advanced Technology; Shanghai AI Laboratory; University of Sydney; The Hong Kong Polytechnic University; ARC Lab, Tencent PCG; The Chinese University of Hong Kong 🚀 We're thrilled to announce the official launch of SupPixel AI! Experience the next level of image processing and upscaling with our cutting-edge AI technology based on SUPIR. Explore now at suppixel.ai. 🔧 Dependencies and Installation Clone repo Install dependent packages Download Checkpoints For users who can connect to huggingface, please setting LLAVACLIPPATH, SDXLCLIP1PATH, SDXLCLIP2CKPTPTH in CKPTPTH.py as None. These CLIPs will be downloaded automatically. Dependent Models SDXL CLIP Encoder-1 SDXL CLIP Encoder-2 SDXL base 1.00.9vae LLaVA CLIP LLaVA v1.5 13B (optional) Juggernaut-XLv9RunDiffusionPhotov2 Replacement of SDXL base 1.0_0.9vae for Photo Realistic (optional) JuggernautRunDiffusionPhoto2Lightning4Steps Distilling model used in SUPIRv0Juggernautv9_lightning.yaml Models we provided: SUPIR-v0Q: Baidu Netdisk, Google Drive Default training settings with paper. High generalization and high image quality in most cases. SUPIR-v0F: Baidu Netdisk, Google Drive Training with light degradation settings. Stage1 encoder of SUPIR-v0F remains more details when facing light degradations. Edit Custom Path for Checkpoints ⚡ Quick Inference Val Dataset RealPhoto60: Baidu Netdisk, Google Drive Usage of SUPIR Python Script Gradio Demo Online App We've just launched SupPixel AI, an easy-to-use tool designed to help with high-quality image processing and upscaling. It builds on SUPIR. Whether you’re into photography, digital art, or just love playing around with image enhancement, we’d love for you to check it out.~ BibTeX @misc{yu2024scaling, title={Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild}, author={Fanghua Yu and Jinjin Gu and Zheyuan Li and Jinfan Hu and Xiangtao Kong and Xintao Wang and Jingwen He and Yu Qiao and Chao Dong}, year={2024}, eprint={2401.13627}, archivePrefix={arXiv}, primaryClass={cs.CV} } 📧 Contact If you have any question, please email fanghuayu96@gmail.com or jinjin.gu@suppixel.ai. Non-Commercial Use Only Declaration The SUPIR ("Software") is made available for use, reproduction, and distribution strictly for non-commercial purposes. For the purposes of this declaration, "non-commercial" is defined as not primarily intended for or directed towards commercial advantage or monetary compensation. By using, reproducing, or distributing the Software, you agree to abide by this restriction and not to use the Software for any commercial purposes without obtaining prior written permission from Dr. Jinjin Gu. This declaration does not in any way limit the rights under any open source license that may apply to the Software; it solely adds a condition that the Software shall not be used for commercial purposes. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. For inquiries or to obtain permission for commercial use, please contact Dr. Jinjin Gu (jinjin.gu@suppixel.ai).

Mintplex-Labs•Mar 28, 2025

anything-llm

AnythingLLM: The all-in-one AI app you were looking for. Chat with your docs, use AI Agents, hyper-configurable, multi-user, & no frustrating set up required. | | Docs | Hosted Instance English · 简体中文 · 日本語 👉 AnythingLLM for desktop (Mac, Windows, & Linux)! Download Now A full-stack application that enables you to turn any document, resource, or piece of content into context that any LLM can use as references during chatting. This application allows you to pick and choose which LLM or Vector Database you want to use as well as supporting multi-user management and permissions. !Chatting Watch the demo! Product Overview AnythingLLM is a full-stack application where you can use commercial off-the-shelf LLMs or popular open source LLMs and vectorDB solutions to build a private ChatGPT with no compromises that you can run locally as well as host remotely and be able to chat intelligently with any documents you provide it. AnythingLLM divides your documents into objects called workspaces. A Workspace functions a lot like a thread, but with the addition of containerization of your documents. Workspaces can share documents, but they do not talk to each other so you can keep your context for each workspace clean. Cool features of AnythingLLM 🆕 Custom AI Agents 🆕 No-code AI Agent builder 🖼️ Multi-modal support (both closed and open-source LLMs!) 👤 Multi-user instance support and permissioning Docker version only 🦾 Agents inside your workspace (browse the web, etc) 💬 Custom Embeddable Chat widget for your website Docker version only 📖 Multiple document type support (PDF, TXT, DOCX, etc) Simple chat UI with Drag-n-Drop funcitonality and clear citations. 100% Cloud deployment ready. Works with all popular closed and open-source LLM providers. Built-in cost & time-saving measures for managing very large documents compared to any other chat UI. Full Developer API for custom integrations! Much more...install and find out! Supported LLMs, Embedder Models, Speech models, and Vector Databases Large Language Models (LLMs): Any open-source llama.cpp compatible model OpenAI OpenAI (Generic) Azure OpenAI AWS Bedrock Anthropic NVIDIA NIM (chat models) Google Gemini Pro Hugging Face (chat models) Ollama (chat models) LM Studio (all models) LocalAi (all models) Together AI (chat models) Fireworks AI (chat models) Perplexity (chat models) OpenRouter (chat models) DeepSeek (chat models) Mistral Groq Cohere KoboldCPP LiteLLM Text Generation Web UI Apipie xAI Novita AI (chat models) PPIO Embedder models: AnythingLLM Native Embedder (default) OpenAI Azure OpenAI LocalAi (all) Ollama (all) LM Studio (all) Cohere Audio Transcription models: AnythingLLM Built-in (default) OpenAI TTS (text-to-speech) support: Native Browser Built-in (default) PiperTTSLocal - runs in browser OpenAI TTS ElevenLabs Any OpenAI Compatible TTS service. STT (speech-to-text) support: Native Browser Built-in (default) Vector Databases: LanceDB (default) Astra DB Pinecone Chroma Weaviate Qdrant Milvus Zilliz Technical Overview This monorepo consists of three main sections: frontend: A viteJS + React frontend that you can run to easily create and manage all your content the LLM can use. server: A NodeJS express server to handle all the interactions and do all the vectorDB management and LLM interactions. collector: NodeJS express server that process and parses documents from the UI. docker: Docker instructions and build process + information for building from source. embed: Submodule for generation & creation of the web embed widget. browser-extension: Submodule for the chrome browser extension. 🛳 Self Hosting Mintplex Labs & the community maintain a number of deployment methods, scripts, and templates that you can use to run AnythingLLM locally. Refer to the table below to read how to deploy on your preferred environment or to automatically deploy. | Docker | AWS | GCP | Digital Ocean | Render.com | |----------------------------------------|----|-----|---------------|------------| | [![Deploy on Docker][docker-btn]][docker-deploy] | [![Deploy on AWS][aws-btn]][aws-deploy] | [![Deploy on GCP][gcp-btn]][gcp-deploy] | [![Deploy on DigitalOcean][do-btn]][do-deploy] | [![Deploy on Render.com][render-btn]][render-deploy] | | Railway | RepoCloud | Elestio | | --- | --- | --- | | [![Deploy on Railway][railway-btn]][railway-deploy] | [![Deploy on RepoCloud][repocloud-btn]][repocloud-deploy] | [![Deploy on Elestio][elestio-btn]][elestio-deploy] | or set up a production AnythingLLM instance without Docker → How to setup for development yarn setup To fill in the required .env files you'll need in each of the application sections (from root of repo). Go fill those out before proceeding. Ensure server/.env.development is filled or else things won't work right. yarn dev:server To boot the server locally (from root of repo). yarn dev:frontend To boot the frontend locally (from root of repo). yarn dev:collector To then run the document collector (from root of repo). Learn about documents Learn about vector caching External Apps & Integrations These are apps that are not maintained by Mintplex Labs, but are compatible with AnythingLLM. A listing here is not an endorsement. Midori AI Subsystem Manager - A streamlined and efficient way to deploy AI systems using Docker container technology. Coolify - Deploy AnythingLLM with a single click. GPTLocalhost for Microsoft Word - A local Word Add-in for you to use AnythingLLM in Microsoft Word. Telemetry & Privacy AnythingLLM by Mintplex Labs Inc contains a telemetry feature that collects anonymous usage information. More about Telemetry & Privacy for AnythingLLM Why? We use this information to help us understand how AnythingLLM is used, to help us prioritize work on new features and bug fixes, and to help us improve AnythingLLM's performance and stability. Opting out Set DISABLE_TELEMETRY in your server or docker .env settings to "true" to opt out of telemetry. You can also do this in-app by going to the sidebar > Privacy and disabling telemetry. What do you explicitly track? We will only track usage details that help us make product and roadmap decisions, specifically: Type of your installation (Docker or Desktop) When a document is added or removed. No information about the document. Just that the event occurred. This gives us an idea of use. Type of vector database in use. Let's us know which vector database provider is the most used to prioritize changes when updates arrive for that provider. Type of LLM in use. Let's us know the most popular choice and prioritize changes when updates arrive for that provider. Chat is sent. This is the most regular "event" and gives us an idea of the daily-activity of this project across all installations. Again, only the event is sent - we have no information on the nature or content of the chat itself. You can verify these claims by finding all locations Telemetry.sendTelemetry is called. Additionally these events are written to the output log so you can also see the specific data which was sent - if enabled. No IP or other identifying information is collected. The Telemetry provider is PostHog - an open-source telemetry collection service. View all telemetry events in source code 👋 Contributing create issue create PR with branch name format of - LGTM from core-team 🌟 Contributors 🔗 More Products [VectorAdmin][vector-admin]: An all-in-one GUI & tool-suite for managing vector databases. [OpenAI Assistant Swarm][assistant-swarm]: Turn your entire library of OpenAI assistants into one single army commanded from a single agent. [![][back-to-top]](#readme-top) Copyright © 2025 [Mintplex Labs][profile-link]. This project is MIT licensed. [back-to-top]: https://img.shields.io/badge/-BACKTOTOP-222628?style=flat-square [profile-link]: https://github.com/mintplex-labs [vector-admin]: https://github.com/mintplex-labs/vector-admin [assistant-swarm]: https://github.com/Mintplex-Labs/openai-assistant-swarm [docker-btn]: ./images/deployBtns/docker.png [docker-deploy]: ./docker/HOWTOUSE_DOCKER.md [aws-btn]: ./images/deployBtns/aws.png [aws-deploy]: ./cloud-deployments/aws/cloudformation/DEPLOY.md [gcp-btn]: https://deploy.cloud.run/button.svg [gcp-deploy]: ./cloud-deployments/gcp/deployment/DEPLOY.md [do-btn]: https://www.deploytodo.com/do-btn-blue.svg [do-deploy]: ./cloud-deployments/digitalocean/terraform/DEPLOY.md [render-btn]: https://render.com/images/deploy-to-render-button.svg [render-deploy]: https://render.com/deploy?repo=https://github.com/Mintplex-Labs/anything-llm&branch=render [render-btn]: https://render.com/images/deploy-to-render-button.svg [render-deploy]: https://render.com/deploy?repo=https://github.com/Mintplex-Labs/anything-llm&branch=render [railway-btn]: https://railway.app/button.svg [railway-deploy]: https://railway.app/template/HNSCS1?referralCode=WFgJkn [repocloud-btn]: https://d16t0pc4846x52.cloudfront.net/deploylobe.svg [repocloud-deploy]: https://repocloud.io/details/?app_id=276 [elestio-btn]: https://elest.io/images/logos/deploy-to-elestio-btn.png [elestio-deploy]: https://elest.io/open-source/anythingllm

#github #anything-llm

github

0.457

0.0831216059433162

xpert-ai•Mar 28, 2025

xpert

English | 中文 [uri_license]: https://www.gnu.org/licenses/agpl-3.0.html [urilicenseimage]: https://img.shields.io/badge/License-AGPL%20v3-blue.svg Xpert Cloud · Self-hosting · Documentation · Enterprise inquiry Open-Source AI Platform for Enterprise Data Analysis, Indicator Management and Agents Orchestration Xpert AI is an open-source enterprise-level AI system that perfectly integrates two major platforms: agent orchestration and data analysis. 💡 What's New Agent and Workflow Hybrid Architecture In today's rapidly evolving AI landscape, enterprises face a critical dilemma: how to balance the creativity of LLMs with the stability of processes? While purely agent-based architectures offer flexibility, they are difficult to control; traditional workflows, though reliable, lack adaptability. The Agent and Workflow Hybrid Architecture of the Xpert AI platform is designed to resolve this conflict — it allows AI to possess "free will" while adhering to "rules and order." !agent-workflow-hybrid-architecture Blog - Agent and Workflow Hybrid Architecture Agent Orchestration Platform By coordinating the collaboration of multiple agents, Xpert completes complex tasks. Xpert integrates different types of AI agents through an efficient management mechanism, utilizing their capabilities to solve multidimensional problems. Xpert Agents Data Analysis Platform An agile data analysis platform based on cloud computing for multidimensional modeling, indicator management, and BI display. It supports connecting to various data sources, achieving efficient and flexible data analysis and visualization, and provides multiple intelligent analysis functions and tools to help enterprises quickly and accurately discover business value and make operational decisions. ChatBI ChatBI is an innovative feature we are introducing, combining chat functionality with business intelligence (BI) analysis capabilities. It offers users a more intuitive and convenient data analysis experience through natural language interaction. ChatBI_Demo.mp4 🚀 Quick Start Before installing Xpert, make sure your machine meets the following minimum system requirements: CPU >= 2 Core RAM >= 4 GiB Node.js (ESM and CommonJS) - 18.x, 19.x, 20.x, 22.x The easiest way to start the Xpert server is through docker compose. Before running Xpert with the following commands, make sure that Docker and Docker Compose are installed on your machine: After running, you can access the Xpert dashboard in your browser at http://localhost/onboarding and start the initialization process. Please check our Wiki - Development to get started quickly. 🎯 Mission Empowering enterprises with intelligent collaboration and data-driven insights through innovative AI orchestration and agile analytics. 🌼 Screenshots Show / Hide Screenshots Pareto analysis open in new tab !Pareto analysis Screenshot Product profit analysis open in new tab !Product profit analysis Screenshot Reseller analysis open in new tab !Reseller analysis Screenshot Bigview dashboard open in new tab !Bigview dashboard Screenshot Indicator application open in new tab !Indicator application Screenshot Indicator mobile app open in new tab !Indicator mobile app Screenshot 💻 Demo, Downloads, Testing and Production Demo Xpert AI Platform Demo at . Notes: You can generate samples data in the home dashbaord page. Production (SaaS) Xpert AI Platform SaaS is available at . Note: it's currently in Alpha version / in testing mode, please use it with caution! 🧱 Technology Stack and Requirements TypeScript language NodeJs / NestJs Nx Angular RxJS TypeORM Langchain ECharts Java Mondrian For Production, we recommend: PostgreSQL PM2 See also README.md and CREDITS.md files in relevant folders for lists of libraries and software included in the Platform, information about licenses, and other details 📄 Documentation Please refer to our official Platform Documentation and to our Wiki (WIP). 💌 Contact Us For business inquiries: Xpert AI Platform @ Twitter 🛡️ License We support the open-source community. This software is available under the following licenses: Xpert AI Platform Community Edition Xpert AI Platform Small Business Xpert AI Platform Enterprise Please see LICENSE for more information on licenses. 💪 Thanks to our Contributors Contributors Please give us :star: on Github, it helps! You are more than welcome to submit feature requests in the Xpert AI repo Pull requests are always welcome! Please base pull requests against the develop branch and follow the contributing guide.

superiorlu•Mar 28, 2025

AITreasureBox

AI TreasureBox English | 中文 Collect practical AI repos, tools, websites, papers and tutorials on AI. Translated from ChatGPT, picture from Midjourney. Catalog Repos Tools Websites Report&Paper Tutorials Repos updated repos and stars every 2 hours and re-ranking automatically. | No. | Repos | Description | | ----:|:-----------------------------------------|:------------------------------------------------------------------------------------------------------| | 1|🔥codecrafters-io/build-your-own-x !2025-03-28364681428|Master programming by recreating your favorite technologies from scratch.| | 2|sindresorhus/awesome !2025-03-28353614145|😎 Awesome lists about all kinds of interesting topics| | 3|public-apis/public-apis !2025-03-28334299125|A collective list of free APIs| | 4|kamranahmedse/developer-roadmap !2025-03-2831269540|Interactive roadmaps, guides and other educational content to help developers grow in their careers.| | 5|vinta/awesome-python !2025-03-28238581114|A curated list of awesome Python frameworks, libraries, software and resources| | 6|practical-tutorials/project-based-learning !2025-03-28222661124|Curated list of project-based tutorials| | 7|tensorflow/tensorflow !2025-03-281888714|An Open Source Machine Learning Framework for Everyone| | 8|Significant-Gravitas/AutoGPT !2025-03-2817391338|An experimental open-source attempt to make GPT-4 fully autonomous.| | 9|jackfrued/Python-100-Days !2025-03-2816305141|Python - 100天从新手到大师| | 10|AUTOMATIC1111/stable-diffusion-webui !2025-03-2815011553|Stable Diffusion web UI| | 11|huggingface/transformers !2025-03-2814207850|🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.| | 12|ollama/ollama !2025-03-28135166151|Get up and running with Llama 2, Mistral, Gemma, and other large language models.| | 13|f/awesome-chatgpt-prompts !2025-03-2812212738 |This repo includes ChatGPT prompt curation to use ChatGPT better.| | 14|justjavac/free-programming-books-zhCN !2025-03-2811316119|📚 免费的计算机编程类中文书籍，欢迎投稿| | 15|krahets/hello-algo !2025-03-2811107930|《Hello 算法》：动画图解、一键运行的数据结构与算法教程。支持 Python, Java, C++, C, C#, JS, Go, Swift, Rust, Ruby, Kotlin, TS, Dart 代码。简体版和繁体版同步更新，English version ongoing| | 16|yt-dlp/yt-dlp !2025-03-28105801114|A feature-rich command-line audio/video downloader| | 17|langchain-ai/langchain !2025-03-2810449479|⚡ Building applications with LLMs through composability ⚡| | 18|goldbergyoni/nodebestpractices !2025-03-281021629|✅ The Node.js best practices list (July 2024)| | 19|puppeteer/puppeteer !2025-03-289018212|JavaScript API for Chrome and Firefox| | 20|pytorch/pytorch !2025-03-288833938|Tensors and Dynamic neural networks in Python with strong GPU acceleration| | 21|neovim/neovim !2025-03-288781482|Vim-fork focused on extensibility and usability| | 22|🔥🔥langgenius/dify !2025-03-2887342639 |One API for plugins and datasets, one interface for prompt engineering and visual operation, all for creating powerful AI applications.| | 23|mtdvio/every-programmer-should-know !2025-03-28867069|A collection of (mostly) technical things every software developer should know about| | 24|open-webui/open-webui !2025-03-2886025159|User-friendly WebUI for LLMs (Formerly Ollama WebUI)| | 25|ChatGPTNextWeb/NextChat !2025-03-288231521|✨ Light and Fast AI Assistant. Support: Web | | 26|supabase/supabase !2025-03-287990956|The open source Firebase alternative.| | 27|openai/whisper !2025-03-287905542|Robust Speech Recognition via Large-Scale Weak Supervision| | 28|home-assistant/core !2025-03-287773219|🏡 Open source home automation that puts local control and privacy first.| | 29|tensorflow/models !2025-03-28774694|Models and examples built with TensorFlow| | 30| ggerganov/llama.cpp !2025-03-287731836 | Port of Facebook's LLaMA model in C/C++ | | 31|3b1b/manim !2025-03-287641918|Animation engine for explanatory math videos| | 32|microsoft/generative-ai-for-beginners !2025-03-287623860|12 Lessons, Get Started Building with Generative AI 🔗 https://microsoft.github.io/generative-ai-for-beginners/| | 33|nomic-ai/gpt4all !2025-03-28729285 |gpt4all: an ecosystem of open-source chatbots trained on a massive collection of clean assistant data including code, stories and dialogue| | 34|comfyanonymous/ComfyUI !2025-03-2872635111|The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.| | 35|bregman-arie/devops-exercises !2025-03-2872225209|Linux, Jenkins, AWS, SRE, Prometheus, Docker, Python, Ansible, Git, Kubernetes, Terraform, OpenStack, SQL, NoSQL, Azure, GCP, DNS, Elastic, Network, Virtualization. DevOps Interview Questions| | 36|elastic/elasticsearch !2025-03-28721419|Free and Open, Distributed, RESTful Search Engine| | 37|🔥n8n-io/n8n !2025-03-2872093495|Free and source-available fair-code licensed workflow automation tool. Easily automate tasks across different services.| | 38|fighting41love/funNLP !2025-03-287200422|The Most Powerful NLP-Weapon Arsenal| | 39|hoppscotch/hoppscotch !2025-03-287060134|Open source API development ecosystem - https://hoppscotch.io (open-source alternative to Postman, Insomnia)| | 40|abi/screenshot-to-code !2025-03-286932817|Drop in a screenshot and convert it to clean HTML/Tailwind/JS code| | 41|binary-husky/gptacademic !2025-03-28680374|Academic Optimization of GPT| | 42|d2l-ai/d2l-zh !2025-03-286774142|Targeting Chinese readers, functional and open for discussion. The Chinese and English versions are used for teaching in over 400 universities across more than 60 countries| | 43|josephmisiti/awesome-machine-learning !2025-03-286739215|A curated list of awesome Machine Learning frameworks, libraries and software.| | 44|grafana/grafana !2025-03-286725414|The open and composable observability and data visualization platform. Visualize metrics, logs, and traces from multiple sources like Prometheus, Loki, Elasticsearch, InfluxDB, Postgres and many more.| | 45|python/cpython !2025-03-286602218|The Python programming language| | 46|apache/superset !2025-03-286519020|Apache Superset is a Data Visualization and Data Exploration Platform| | 47|xtekky/gpt4free !2025-03-28639391 |decentralizing the Ai Industry, free gpt-4/3.5 scripts through several reverse engineered API's ( poe.com, phind.com, chat.openai.com etc...)| | 48|sherlock-project/sherlock !2025-03-286332536|Hunt down social media accounts by username across social networks| | 49|twitter/the-algorithm !2025-03-28630586 |Source code for Twitter's Recommendation Algorithm| | 50|keras-team/keras !2025-03-28627835|Deep Learning for humans| | 51|openai/openai-cookbook !2025-03-28625136 |Examples and guides for using the OpenAI API| | 52|immich-app/immich !2025-03-286238670|High performance self-hosted photo and video management solution.| | 53|AppFlowy-IO/AppFlowy !2025-03-286173528|Bring projects, wikis, and teams together with AI. AppFlowy is an AI collaborative workspace where you achieve more without losing control of your data. The best open source alternative to Notion.| | 54|scikit-learn/scikit-learn !2025-03-286158212|scikit-learn: machine learning in Python| | 55|binhnguyennus/awesome-scalability !2025-03-286117021|The Patterns of Scalable, Reliable, and Performant Large-Scale Systems| | 56|labmlai/annotateddeeplearningpaperimplementations !2025-03-285951726|🧑‍🏫 59 Implementations/tutorials of deep learning papers with side-by-side notes 📝; including transformers (original, xl, switch, feedback, vit, ...), optimizers (adam, adabelief, ...), gans(cyclegan, stylegan2, ...), 🎮 reinforcement learning (ppo, dqn), capsnet, distillation, ... 🧠| | 57|OpenInterpreter/open-interpreter !2025-03-285894710|A natural language interface for computers| | 58|lobehub/lobe-chat !2025-03-285832054|🤖 Lobe Chat - an open-source, extensible (Function Calling), high-performance chatbot framework. It supports one-click free deployment of your private ChatGPT/LLM web application.| | 59|meta-llama/llama !2025-03-28579536|Inference code for Llama models| | 60|nuxt/nuxt !2025-03-28566437|The Intuitive Vue Framework.| | 61|imartinez/privateGPT !2025-03-28555192|Interact with your documents using the power of GPT, 100% privately, no data leaks| | 62|Stirling-Tools/Stirling-PDF !2025-03-285500846|#1 Locally hosted web application that allows you to perform various operations on PDF files| | 63|PlexPt/awesome-chatgpt-prompts-zh !2025-03-285459720|ChatGPT Chinese Training Guide. Guidelines for various scenarios. Learn how to make it listen to you| | 64|dair-ai/Prompt-Engineering-Guide !2025-03-285451025 |🐙 Guides, papers, lecture, notebooks and resources for prompt engineering| | 65|ageitgey/facerecognition !2025-03-28544382|The world's simplest facial recognition api for Python and the command line| | 66|CorentinJ/Real-Time-Voice-Cloning !2025-03-285384814|Clone a voice in 5 seconds to generate arbitrary speech in real-time| | 67|geekan/MetaGPT !2025-03-285375376|The Multi-Agent Meta Programming Framework: Given one line Requirement, return PRD, Design, Tasks, Repo | | 68|gpt-engineer-org/gpt-engineer !2025-03-285367419|Specify what you want it to build, the AI asks for clarification, and then builds it.| | 69|lencx/ChatGPT !2025-03-2853653-3|🔮 ChatGPT Desktop Application (Mac, Windows and Linux)| | 70|deepfakes/faceswap !2025-03-28535672|Deepfakes Software For All| | 71|langflow-ai/langflow !2025-03-285319584|Langflow is a low-code app builder for RAG and multi-agent AI applications. It’s Python-based and agnostic to any model, API, or database.| | 72|commaai/openpilot !2025-03-28529759|openpilot is an operating system for robotics. Currently, it upgrades the driver assistance system on 275+ supported cars.| | 73|clash-verge-rev/clash-verge-rev !2025-03-2852848124|Continuation of Clash Verge - A Clash Meta GUI based on Tauri (Windows, MacOS, Linux)| | 74|All-Hands-AI/OpenHands !2025-03-285150675|🙌 OpenHands: Code Less, Make More| | 75|xai-org/grok-1 !2025-03-28502504|Grok open release| | 76|meilisearch/meilisearch !2025-03-284999122|A lightning-fast search API that fits effortlessly into your apps, websites, and workflow| | 77|🔥browser-use/browser-use !2025-03-2849910294|Make websites accessible for AI agents| | 78|jgthms/bulma !2025-03-28496783|Modern CSS framework based on Flexbox| | 79|facebookresearch/segment-anything !2025-03-284947116|The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.| |!green-up-arrow.svg 80|hacksider/Deep-Live-Cam !2025-03-2848612146|real time face swap and one-click video deepfake with only a single image (uncensored)| |!red-down-arrow 81|mlabonne/llm-course !2025-03-284860934|Course with a roadmap and notebooks to get into Large Language Models (LLMs).| | 82|PaddlePaddle/PaddleOCR !2025-03-284785530|Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)| | 83|alist-org/alist !2025-03-284732618|🗂️A file list/WebDAV program that supports multiple storages, powered by Gin and Solidjs. / 一个支持多存储的文件列表/WebDAV程序，使用 Gin 和 Solidjs。| | 84|infiniflow/ragflow !2025-03-2847027129|RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.| | 85|Avik-Jain/100-Days-Of-ML-Code !2025-03-284679312|100 Days of ML Coding| | 86|v2ray/v2ray-core !2025-03-28458706|A platform for building proxies to bypass network restrictions.| | 87|hiyouga/LLaMA-Factory !2025-03-284555881|Easy-to-use LLM fine-tuning framework (LLaMA, BLOOM, Mistral, Baichuan, Qwen, ChatGLM)| | 88|Asabeneh/30-Days-Of-Python !2025-03-284544930|30 days of Python programming challenge is a step-by-step guide to learn the Python programming language in 30 days. This challenge may take more than100 days, follow your own pace. These videos may help too: https://www.youtube.com/channel/UC7PNRuno1rzYPb1xLa4yktw| | 89|type-challenges/type-challenges !2025-03-284488511|Collection of TypeScript type challenges with online judge| | 90|lllyasviel/Fooocus !2025-03-284402716|Focus on prompting and generating| | 91|RVC-Boss/GPT-SoVITS !2025-03-284327738|1 min voice data can also be used to train a good TTS model! (few shot voice cloning)| | 92|rasbt/LLMs-from-scratch !2025-03-284320667|Implementing a ChatGPT-like LLM from scratch, step by step| | 93|oobabooga/text-generation-webui !2025-03-284302012 |A gradio web UI for running Large Language Models like LLaMA, llama.cpp, GPT-J, OPT, and GALACTICA.| | 94|vllm-project/vllm !2025-03-2842982102|A high-throughput and memory-efficient inference and serving engine for LLMs| | 95|dani-garcia/vaultwarden !2025-03-284297121|Unofficial Bitwarden compatible server written in Rust, formerly known as bitwarden_rs| | 96|microsoft/autogen !2025-03-284233049|Enable Next-Gen Large Language Model Applications. Join our Discord: https://discord.gg/pAbnFJrkgZ| | 97|jeecgboot/JeecgBoot !2025-03-284205920|🔥「企业级低代码平台」前后端分离架构SpringBoot 2.x/3.x，SpringCloud，Ant Design&Vue3，Mybatis，Shiro，JWT。强大的代码生成器让前后端代码一键生成，无需写任何代码! 引领新的开发模式OnlineCoding->代码生成->手工MERGE，帮助Java项目解决70%重复工作，让开发更关注业务，既能快速提高效率，帮助公司节省成本，同时又不失灵活性。| | 98|Mintplex-Labs/anything-llm !2025-03-284186955|A full-stack application that turns any documents into an intelligent chatbot with a sleek UI and easier way to manage your workspaces.| | 99|THUDM/ChatGLM-6B !2025-03-28410192 |ChatGLM-6B: An Open Bilingual Dialogue Language Model| | 100|hpcaitech/ColossalAI !2025-03-28406902|Making large AI models cheaper, faster and more accessible| | 101|Stability-AI/stablediffusion !2025-03-28406337|High-Resolution Image Synthesis with Latent Diffusion Models| | 102|mingrammer/diagrams !2025-03-28405063|🎨 Diagram as Code for prototyping cloud system architectures| | 103|Kong/kong !2025-03-28404616|🦍 The Cloud-Native API Gateway and AI Gateway.| | 104|getsentry/sentry !2025-03-284040913|Developer-first error tracking and performance monitoring| | 105| karpathy/nanoGPT !2025-03-284034613 |The simplest, fastest repository for training/finetuning medium-sized GPTs| | 106|fastlane/fastlane !2025-03-2840014-1|🚀 The easiest way to automate building and releasing your iOS and Android apps| | 107|psf/black !2025-03-28399765|The uncompromising Python code formatter| | 108|OpenBB-finance/OpenBBTerminal !2025-03-283972074 |Investment Research for Everyone, Anywhere.| | 109|2dust/v2rayNG !2025-03-283943415|A V2Ray client for Android, support Xray core and v2fly core| | 110|apache/airflow !2025-03-283937314|Apache Airflow - A platform to programmatically author, schedule, and monitor workflows| | 111|KRTirtho/spotube !2025-03-283902746|🎧 Open source Spotify client that doesn't require Premium nor uses Electron! Available for both desktop & mobile!| | 112|coqui-ai/TTS !2025-03-283889719 |🐸💬 - a deep learning toolkit for Text-to-Speech, battle-tested in research and production| | 113|ggerganov/whisper.cpp !2025-03-283882116|Port of OpenAI's Whisper model in C/C++| | 114|ultralytics/ultralytics !2025-03-283866951|NEW - YOLOv8 🚀 in PyTorch > ONNX > OpenVINO > CoreML > TFLite| | 115|typst/typst !2025-03-283863914|A new markup-based typesetting system that is powerful and easy to learn.| | 116|streamlit/streamlit !2025-03-283845828|Streamlit — A faster way to build and share data apps.| | 117|LC044/WeChatMsg !2025-03-283836931|提取微信聊天记录，将其导出成HTML、Word、Excel文档永久保存，对聊天记录进行分析生成年度聊天报告，用聊天数据训练专属于个人的AI聊天助手| | 118|lm-sys/FastChat !2025-03-283822112 |An open platform for training, serving, and evaluating large languages. Release repo for Vicuna and FastChat-T5.| | 119|NaiboWang/EasySpider !2025-03-283819013|A visual no-code/code-free web crawler/spider易采集：一个可视化浏览器自动化测试/数据采集/爬虫软件，可以无代码图形化的设计和执行爬虫任务。别名：ServiceWrapper面向Web应用的智能化服务封装系统。| | 120|microsoft/DeepSpeed !2025-03-283765816 |A deep learning optimization library that makes distributed training and inference easy, efficient, and effective| | 121|QuivrHQ/quivr !2025-03-28376067|Your GenAI Second Brain 🧠 A personal productivity assistant (RAG) ⚡️🤖 Chat with your docs (PDF, CSV, ...) & apps using Langchain, GPT 3.5 / 4 turbo, Private, Anthropic, VertexAI, Ollama, LLMs, that you can share with users ! Local & Private alternative to OpenAI GPTs & ChatGPT powered by retrieval-augmented generation.| | 122|freqtrade/freqtrade !2025-03-283757817 |Free, open source crypto trading bot| | 123|suno-ai/bark !2025-03-28373178 |🔊 Text-Prompted Generative Audio Model| | 124|🔥cline/cline !2025-03-2837307282|Autonomous coding agent right in your IDE, capable of creating/editing files, executing commands, and more with your permission every step of the way.| | 125|LAION-AI/Open-Assistant !2025-03-28372712 |OpenAssistant is a chat-based assistant that understands tasks, can interact with third-party systems, and retrieve information dynamically to do so.| | 126|penpot/penpot !2025-03-283716217|Penpot: The open-source design tool for design and code collaboration| | 127|gradio-app/gradio !2025-03-283713320|Build and share delightful machine learning apps, all in Python. 🌟 Star to support our work!| | 128|FlowiseAI/Flowise !2025-03-283667135 |Drag & drop UI to build your customized LLM flow using LangchainJS| | 129|SimplifyJobs/Summer2025-Internships !2025-03-28366506|Collection of Summer 2025 tech internships!| | 130|TencentARC/GFPGAN !2025-03-28365027 |GFPGAN aims at developing Practical Algorithms for Real-world Face Restoration.| | 131|ray-project/ray !2025-03-283626819|Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a toolkit of libraries (Ray AIR) for accelerating ML workloads.| | 132|babysor/MockingBird !2025-03-28360498|🚀AI拟声: 5秒内克隆您的声音并生成任意语音内容 Clone a voice in 5 seconds to generate arbitrary speech in real-time| | 133|unslothai/unsloth !2025-03-283603691|5X faster 50% less memory LLM finetuning| | 134|zhayujie/chatgpt-on-wechat !2025-03-283600124 |Wechat robot based on ChatGPT, which uses OpenAI api and itchat library| | 135|upscayl/upscayl !2025-03-283599824|🆙 Upscayl - Free and Open Source AI Image Upscaler for Linux, MacOS and Windows built with Linux-First philosophy.| | 136|freeCodeCamp/devdocs !2025-03-28359738|API Documentation Browser| | 137|XingangPan/DragGAN !2025-03-28359043 |Code for DragGAN (SIGGRAPH 2023)| | 138|2noise/ChatTTS !2025-03-283543922|ChatTTS is a generative speech model for daily dialogue.| | 139|google-research/google-research !2025-03-28352207 |Google Research| | 140|karanpratapsingh/system-design !2025-03-28351003|Learn how to design systems at scale and prepare for system design interviews| | 141|lapce/lapce !2025-03-28350855|Lightning-fast and Powerful Code Editor written in Rust| | 142| microsoft/TaskMatrix !2025-03-2834500-3 | Talking, Drawing and Editing with Visual Foundation Models| | 143|chatchat-space/Langchain-Chatchat !2025-03-283442020|Langchain-Chatchat (formerly langchain-ChatGLM), local knowledge based LLM (like ChatGLM) QA app with langchain| | 144|unclecode/crawl4ai !2025-03-283434163|🔥🕷️ Crawl4AI: Open-source LLM Friendly Web Crawler & Scrapper| | 145|Bin-Huang/chatbox !2025-03-283374733 |A desktop app for GPT-4 / GPT-3.5 (OpenAI API) that supports Windows, Mac & Linux| | 146|milvus-io/milvus !2025-03-283366525 |A cloud-native vector database, storage for next generation AI applications| | 147|mendableai/firecrawl !2025-03-2833297128|🔥 Turn entire websites into LLM-ready markdown| | 148|pola-rs/polars !2025-03-283269320|Fast multi-threaded, hybrid-out-of-core query engine focussing on DataFrame front-ends| | 149|Pythagora-io/gpt-pilot !2025-03-28325321|PoC for a scalable dev tool that writes entire apps from scratch while the developer oversees the implementation| | 150|hashicorp/vault !2025-03-28320797|A tool for secrets management, encryption as a service, and privileged access management| | 151|shardeum/shardeum !2025-03-28319580|Shardeum is an EVM based autoscaling blockchain| | 152|Chanzhaoyu/chatgpt-web !2025-03-28319242 |A demonstration website built with Express and Vue3 called ChatGPT| | 153|lllyasviel/ControlNet !2025-03-283186413 |Let us control diffusion models!| | 154|google/jax !2025-03-28317727|Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more| | 155|facebookresearch/detectron2 !2025-03-28315987|Detectron2 is a platform for object detection, segmentation and other visual recognition tasks.| | 156|myshell-ai/OpenVoice !2025-03-28315233|Instant voice cloning by MyShell| | 157|TheAlgorithms/C-Plus-Plus !2025-03-283151411|Collection of various algorithms in mathematics, machine learning, computer science and physics implemented in C++ for educational purposes.| | 158|hiroi-sora/Umi-OCR !2025-03-283138129|OCR图片转文字识别软件，完全离线。截屏/批量导入图片，支持多国语言、合并段落、竖排文字。可排除水印区域，提取干净的文本。基于 PaddleOCR 。| | 159|mudler/LocalAI !2025-03-283127815|🤖 The free, Open Source OpenAI alternative. Self-hosted, community-driven and local-first. Drop-in replacement for OpenAI running on consumer-grade hardware. No GPU required. Runs gguf, transformers, diffusers and many more models architectures. It allows to generate Text, Audio, Video, Images. Also with voice cloning capabilities.| | 160|facebookresearch/fairseq !2025-03-28312124 |Facebook AI Research Sequence-to-Sequence Toolkit written in Python.| | 161|alibaba/nacos !2025-03-28310559|an easy-to-use dynamic service discovery, configuration and service management platform for building cloud native applications.| | 162|yunjey/pytorch-tutorial !2025-03-28310326|PyTorch Tutorial for Deep Learning Researchers| | 163|v2fly/v2ray-core !2025-03-28307448|A platform for building proxies to bypass network restrictions.| | 164|mckaywrigley/chatbot-ui !2025-03-283067714|The open-source AI chat interface for everyone.| | 165|TabbyML/tabby !2025-03-28305949 |Self-hosted AI coding assistant| | 166|deepseek-ai/awesome-deepseek-integration !2025-03-283053193|| | 167|danielmiessler/fabric !2025-03-283028914|fabric is an open-source framework for augmenting humans using AI.| | 168|xinntao/Real-ESRGAN !2025-03-283026623 |Real-ESRGAN aims at developing Practical Algorithms for General Image/Video Restoration.| | 169|paul-gauthier/aider !2025-03-283014642|aider is GPT powered coding in your terminal| | 170|tatsu-lab/stanfordalpaca !2025-03-28299022 |Code and documentation to train Stanford's Alpaca models, and generate the data.| | 171|DataTalksClub/data-engineering-zoomcamp !2025-03-282971817|Free Data Engineering course!| | 172|HeyPuter/puter !2025-03-282967014|🌐 The Internet OS! Free, Open-Source, and Self-Hostable.| | 173|mli/paper-reading !2025-03-282962314|Classic Deep Learning and In-Depth Reading of New Papers Paragraph by Paragraph| | 174|linexjlin/GPTs !2025-03-28295568|leaked prompts of GPTs| | 175|s0md3v/roop !2025-03-28295286 |one-click deepfake (face swap)| | 176|JushBJJ/Mr.-Ranedeer-AI-Tutor !2025-03-2829465-1 |A GPT-4 AI Tutor Prompt for customizable personalized learning experiences.| | 177|opendatalab/MinerU !2025-03-282927074|A one-stop, open-source, high-quality data extraction tool, supports PDF/webpage/e-book extraction.一站式开源高质量数据提取工具，支持PDF/网页/多格式电子书提取。| | 178|mouredev/Hello-Python !2025-03-282920720|Curso para aprender el lenguaje de programación Python desde cero y para principiantes. 75 clases, 37 horas en vídeo, código, proyectos y grupo de chat. Fundamentos, frontend, backend, testing, IA...| | 179|Lightning-AI/pytorch-lightning !2025-03-28292039|Pretrain, finetune and deploy AI models on multiple GPUs, TPUs with zero code changes.| | 180|crewAIInc/crewAI !2025-03-282919344|Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.| | 181|facebook/folly !2025-03-282916612|An open-source C++ library developed and used at Facebook.| | 182|google-ai-edge/mediapipe !2025-03-28291519|Cross-platform, customizable ML solutions for live and streaming media.| | 183| getcursor/cursor !2025-03-282892025 | An editor made for programming with AI| | 184|chatanywhere/GPTAPIfree !2025-03-282856424|Free ChatGPT API Key, Free ChatGPT API, supports GPT-4 API (free), ChatGPT offers a free domestic forwarding API that allows direct connections without the need for a proxy. It can be used in conjunction with software/plugins like ChatBox, significantly reducing interface usage costs. Enjoy unlimited and unrestricted chatting within China| | 185|meta-llama/llama3 !2025-03-28285552|The official Meta Llama 3 GitHub site| | 186|tinygrad/tinygrad !2025-03-282845811|You like pytorch? You like micrograd? You love tinygrad! ❤️| | 187|google-research/tuningplaybook !2025-03-282841514|A playbook for systematically maximizing the performance of deep learning models.| | 188|huggingface/diffusers !2025-03-282830222|🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.| | 189|tokio-rs/tokio !2025-03-28282408|A runtime for writing reliable asynchronous applications with Rust. Provides I/O, networking, scheduling, timers, ...| | 190|RVC-Project/Retrieval-based-Voice-Conversion-WebUI !2025-03-282823817|Voice data !2025-03-282822612|Jan is an open source alternative to ChatGPT that runs 100% offline on your computer| | 192|openai/CLIP !2025-03-282814720|CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image| | 193|🔥khoj-ai/khoj !2025-03-2828112313|Your AI second brain. A copilot to get answers to your questions, whether they be from your own notes or from the internet. Use powerful, online (e.g gpt4) or private, local (e.g mistral) LLMs. Self-host locally or use our web app. Access from Obsidian, Emacs, Desktop app, Web or Whatsapp.| | 194| acheong08/ChatGPT !2025-03-2828054-2 | Reverse engineered ChatGPT API | | 195|iperov/DeepFaceLive !2025-03-28279345 |Real-time face swap for PC streaming or video calls| | 196|eugeneyan/applied-ml !2025-03-28278471|📚 Papers & tech blogs by companies sharing their work on data science & machine learning in production.| | 197|XTLS/Xray-core !2025-03-282778213|Xray, Penetrates Everything. Also the best v2ray-core, with XTLS support. Fully compatible configuration.| | 198|feder-cr/JobsApplierAIAgent !2025-03-282776410|AutoJobsApplierAI_Agent aims to easy job hunt process by automating the job application process. Utilizing artificial intelligence, it enables users to apply for multiple jobs in an automated and personalized way.| | 199|mindsdb/mindsdb !2025-03-282750631|The platform for customizing AI from enterprise data| | 200|DataExpert-io/data-engineer-handbook !2025-03-282721611|This is a repo with links to everything you'd ever want to learn about data engineering| | 201|exo-explore/exo !2025-03-282721633|Run your own AI cluster at home with everyday devices 📱💻 🖥️⌚| | 202|taichi-dev/taichi !2025-03-2826926-1|Productive, portable, and performant GPU programming in Python.| | 203|mem0ai/mem0 !2025-03-282689134|The memory layer for Personalized AI| | 204|svc-develop-team/so-vits-svc !2025-03-28268096 |SoftVC VITS Singing Voice Conversion| | 205|OpenBMB/ChatDev !2025-03-28265624|Create Customized Software using Natural Language Idea (through Multi-Agent Collaboration)| | 206|roboflow/supervision !2025-03-282632010|We write your reusable computer vision tools. 💜| | 207|drawdb-io/drawdb !2025-03-282626913|Free, simple, and intuitive online database design tool and SQL generator.| | 208|karpathy/llm.c !2025-03-28261633|LLM training in simple, raw C/CUDA| | 209|airbnb/lottie-ios !2025-03-28261431|An iOS library to natively render After Effects vector animations| | 210|openai/openai-python !2025-03-282607713|The OpenAI Python library provides convenient access to the OpenAI API from applications written in the Python language.| | 211|academic/awesome-datascience !2025-03-28259876|📝 An awesome Data Science repository to learn and apply for real world problems.| | 212|harry0703/MoneyPrinterTurbo !2025-03-282576618|Generate short videos with one click using a large model| | 213|gabime/spdlog !2025-03-282571511|Fast C++ logging library.| | 214|ocrmypdf/OCRmyPDF !2025-03-2825674217|OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched| | 215|Vision-CAIR/MiniGPT-4 !2025-03-28256170 |Enhancing Vision-language Understanding with Advanced Large Language Models| | 216|Stability-AI/generative-models !2025-03-28255936|Generative Models by Stability AI| | 217|DS4SD/docling !2025-03-282555662|Get your docs ready for gen AI| | 218|PostHog/posthog !2025-03-282533227|🦔 PostHog provides open-source product analytics, session recording, feature flagging and A/B testing that you can self-host.| | 219|nrwl/nx !2025-03-282509612|Smart Monorepos · Fast CI| | 220|continuedev/continue !2025-03-282500737|⏩ the open-source copilot chat for software development—bring the power of ChatGPT to VS Code| | 221|opentofu/opentofu !2025-03-28247968|OpenTofu lets you declaratively manage your cloud infrastructure.| | 222|invoke-ai/InvokeAI !2025-03-28247293|InvokeAI is a leading creative engine for Stable Diffusion models, empowering professionals, artists, and enthusiasts to generate and create visual media using the latest AI-driven technologies. The solution offers an industry leading WebUI, supports terminal use through a CLI, and serves as the foundation for multiple commercial products.| | 223|deepinsight/insightface !2025-03-282471615 |State-of-the-art 2D and 3D Face Analysis Project| | 224|apache/flink !2025-03-28246865|Apache Flink| | 225|ComposioHQ/composio !2025-03-28246436|Composio equips agents with well-crafted tools empowering them to tackle complex tasks| | 226|Genesis-Embodied-AI/Genesis !2025-03-282458314|A generative world for general-purpose robotics & embodied AI learning.| | 227|stretchr/testify !2025-03-28243184|A toolkit with common assertions and mocks that plays nicely with the standard library| | 228| yetone/openai-translator !2025-03-28242921 | Browser extension and cross-platform desktop application for translation based on ChatGPT API | | 229|frappe/erpnext !2025-03-282425211|Free and Open Source Enterprise Resource Planning (ERP)| | 230|songquanpeng/one-api !2025-03-282410034|OpenAI 接口管理 & 分发系统，支持 Azure、Anthropic Claude、Google PaLM 2 & Gemini、智谱 ChatGLM、百度文心一言、讯飞星火认知、阿里通义千问、360 智脑以及腾讯混元，可用于二次分发管理 key，仅单可执行文件，已打包好 Docker 镜像，一键部署，开箱即用. OpenAI key management & redistribution system, using a single API for all LLMs, and features an English UI.| | 231| microsoft/JARVIS !2025-03-28240604 | a system to connect LLMs with ML community | | 232|google/flatbuffers !2025-03-28239965|FlatBuffers: Memory Efficient Serialization Library| | 233|microsoft/graphrag !2025-03-282398928|A modular graph-based Retrieval-Augmented Generation (RAG) system| | 234|rancher/rancher !2025-03-28239675|Complete container management platform| | 235|bazelbuild/bazel !2025-03-282384618|a fast, scalable, multi-language and extensible build system| | 236|modularml/mojo !2025-03-28238236 |The Mojo Programming Language| | 237|danny-avila/LibreChat !2025-03-282378753|Enhanced ChatGPT Clone: Features OpenAI, GPT-4 Vision, Bing, Anthropic, OpenRouter, Google Gemini, AI model switching, message search, langchain, DALL-E-3, ChatGPT Plugins, OpenAI Functions, Secure Multi-User System, Presets, completely open-source for self-hosting. More features in development| |!green-up-arrow.svg 238|🔥🔥🔥Shubhamsaboo/awesome-llm-apps !2025-03-28237391211|Collection of awesome LLM apps with RAG using OpenAI, Anthropic, Gemini and opensource models.| |!red-down-arrow 239|microsoft/semantic-kernel !2025-03-282373611|Integrate cutting-edge LLM technology quickly and easily into your apps| |!red-down-arrow 240|TheAlgorithms/Rust !2025-03-28236995|All Algorithms implemented in Rust| | 241|stanford-oval/storm !2025-03-28236326|An LLM-powered knowledge curation system that researches a topic and generates a full-length report with citations.| | 242|openai/gpt-2 !2025-03-28232483|Code for the paper "Language Models are Unsupervised Multitask Learners"| | 243|labring/FastGPT !2025-03-282319445|A platform that uses the OpenAI API to quickly build an AI knowledge base, supporting many-to-many relationships.| | 244|pathwaycom/llm-app !2025-03-2822928-10|Ready-to-run cloud templates for RAG, AI pipelines, and enterprise search with live data. 🐳Docker-friendly.⚡Always in sync with Sharepoint, Google Drive, S3, Kafka, PostgreSQL, real-time data APIs, and more.| | 245|warpdotdev/Warp !2025-03-282286825|Warp is a modern, Rust-based terminal with AI built in so you and your team can build great software, faster.| | 246|🔥agno-agi/agno !2025-03-2822833298|Agno is a lightweight library for building Multimodal Agents. It exposes LLMs as a unified API and gives them superpowers like memory, knowledge, tools and reasoning.| | 247|qdrant/qdrant !2025-03-282275214 |Qdrant - Vector Database for the next generation of AI applications. Also available in the cloud https://cloud.qdrant.io/| | 248|ashishpatel26/500-AI-Machine-learning-Deep-learning-Computer-vision-NLP-Projects-with-code !2025-03-282271815|500 AI Machine learning Deep learning Computer vision NLP Projects with code| | 249|stanfordnlp/dspy !2025-03-282268321|Stanford DSPy: The framework for programming—not prompting—foundation models| | 250|PaddlePaddle/Paddle !2025-03-28226246|PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）| | 251|zulip/zulip !2025-03-28225464|Zulip server and web application. Open-source team chat that helps teams stay productive and focused.| | 252|Hannibal046/Awesome-LLM !2025-03-282240721|Awesome-LLM: a curated list of Large Language Model| | 253|facefusion/facefusion !2025-03-282218812|Next generation face swapper and enhancer| | 254|Mozilla-Ocho/llamafile !2025-03-28220624|Distribute and run LLMs with a single file.| | 255|yuliskov/SmartTube !2025-03-282201614|SmartTube - an advanced player for set-top boxes and tvs running Android OS| | 256|haotian-liu/LLaVA !2025-03-282201316 |Large Language-and-Vision Assistant built towards multimodal GPT-4 level capabilities.| | 257|ashishps1/awesome-system-design-resources !2025-03-282189367|This repository contains System Design resources which are useful while preparing for interviews and learning Distributed Systems| | 258|Cinnamon/kotaemon !2025-03-28218248|An open-source RAG-based tool for chatting with your documents.| | 259|CodePhiliaX/Chat2DB !2025-03-282179757|🔥🔥🔥AI-driven database tool and SQL client, The hottest GUI client, supporting MySQL, Oracle, PostgreSQL, DB2, SQL Server, DB2, SQLite, H2, ClickHouse, and more.| | 260|blakeblackshear/frigate !2025-03-282177113|NVR with realtime local object detection for IP cameras| | 261|facebookresearch/audiocraft !2025-03-28217111|Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning.| | 262|karpathy/minGPT !2025-03-28216567|A minimal PyTorch re-implementation of the OpenAI GPT (Generative Pretrained Transformer) training| | 263|grpc/grpc-go !2025-03-282159510|The Go language implementation of gRPC. HTTP/2 based RPC| | 264|HumanSignal/label-studio !2025-03-282137618|Label Studio is a multi-type data labeling and annotation tool with standardized output format| | 265|yoheinakajima/babyagi !2025-03-28212764 |uses OpenAI and Pinecone APIs to create, prioritize, and execute tasks, This is a pared-down version of the original Task-Driven Autonomous Agent| | 266|deepseek-ai/DeepSeek-Coder !2025-03-282118210|DeepSeek Coder: Let the Code Write Itself| | 267|BuilderIO/gpt-crawler !2025-03-282118010|Crawl a site to generate knowledge files to create your own custom GPT from a URL| | 268| openai/chatgpt-retrieval-plugin !2025-03-2821152-1 | Plugins are chat extensions designed specifically for language models like ChatGPT, enabling them to access up-to-date information, run computations, or interact with third-party services in response to a user's request.| | 269|microsoft/OmniParser !2025-03-282113123|A simple screen parsing tool towards pure vision based GUI agent| | 270|black-forest-labs/flux !2025-03-282107219|Official inference repo for FLUX.1 models| | 271|ItzCrazyKns/Perplexica !2025-03-282099154|Perplexica is an AI-powered search engine. It is an Open source alternative to Perplexity AI| | 272|microsoft/unilm !2025-03-28209876|Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities| | 273|Sanster/lama-cleaner !2025-03-282077614|Image inpainting tool powered by SOTA AI Model. Remove any unwanted object, defect, people from your pictures or erase and replace(powered by stable diffusion) any thing on your pictures.| | 274|assafelovic/gpt-researcher !2025-03-282057222|GPT based autonomous agent that does online comprehensive research on any given topic| | 275|PromtEngineer/localGPT !2025-03-28204230 |Chat with your documents on your local device using GPT models. No data leaves your device and 100% private.| | 276|elastic/kibana !2025-03-28203482|Your window into the Elastic Stack| | 277|fishaudio/fish-speech !2025-03-282033222|Brand new TTS solution| | 278|mlc-ai/mlc-llm !2025-03-282028110 |Enable everyone to develop, optimize and deploy AI models natively on everyone's devices.| | 279|deepset-ai/haystack !2025-03-282005320|🔍 Haystack is an open source NLP framework to interact with your data using Transformer models and LLMs (GPT-4, ChatGPT and alike). Haystack offers production-ready tools to quickly build complex question answering, semantic search, text generation applications, and more.| | 280|tree-sitter/tree-sitter !2025-03-28200487|An incremental parsing system for programming tools| | 281|Anjok07/ultimatevocalremovergui !2025-03-281999811|GUI for a Vocal Remover that uses Deep Neural Networks.| | 282|guidance-ai/guidance !2025-03-28199622|A guidance language for controlling large language models.| | 283|ml-explore/mlx !2025-03-28199619|MLX: An array framework for Apple silicon| | 284|mlflow/mlflow !2025-03-281995314|Open source platform for the machine learning lifecycle| | 285|ml-tooling/best-of-ml-python !2025-03-28198631|🏆 A ranked list of awesome machine learning Python libraries. Updated weekly.| | 286|BerriAI/litellm !2025-03-281981862|Call all LLM APIs using the OpenAI format. Use Bedrock, Azure, OpenAI, Cohere, Anthropic, Ollama, Sagemaker, HuggingFace, Replicate (100+ LLMs)| | 287|LazyVim/LazyVim !2025-03-281981320|Neovim config for the lazy| | 288|wez/wezterm !2025-03-281976018|A GPU-accelerated cross-platform terminal emulator and multiplexer written by @wez and implemented in Rust| | 289|valkey-io/valkey !2025-03-281970416|A flexible distributed key-value datastore that supports both caching and beyond caching workloads.| | 290|LiLittleCat/awesome-free-chatgpt !2025-03-28196185|🆓免费的 ChatGPT 镜像网站列表，持续更新。List of free ChatGPT mirror sites, continuously updated.| | 291|Byaidu/PDFMathTranslate !2025-03-281947645|PDF scientific paper translation with preserved formats - 基于 AI 完整保留排版的 PDF 文档全文双语翻译，支持 Google/DeepL/Ollama/OpenAI 等服务，提供 CLI/GUI/Docker| | 292|openai/swarm !2025-03-281947111|Educational framework exploring ergonomic, lightweight multi-agent orchestration. Managed by OpenAI Solution team.| | 293|HqWu-HITCS/Awesome-Chinese-LLM !2025-03-281921423|Organizing smaller, cost-effective, privately deployable open-source Chinese language models, including related datasets and tutorials| | 294|stitionai/devika !2025-03-28190903|Devika is an Agentic AI Software Engineer that can understand high-level human instructions, break them down into steps, research relevant information, and write code to achieve the given objective. Devika aims to be a competitive open-source alternative to Devin by Cognition AI.| | 295|OpenBMB/MiniCPM-o !2025-03-28190887|MiniCPM-o 2.6: A GPT-4o Level MLLM for Vision, Speech and Multimodal Live Streaming on Your Phone| | 296|samber/lo !2025-03-281904815|💥 A Lodash-style Go library based on Go 1.18+ Generics (map, filter, contains, find...)| | 297|chroma-core/chroma !2025-03-281895221 |the AI-native open-source embedding database| | 298|DarkFlippers/unleashed-firmware !2025-03-28189278|Flipper Zero Unleashed Firmware| | 299|brave/brave-browser !2025-03-281892710|Brave browser for Android, iOS, Linux, macOS, Windows.| | 300| tloen/alpaca-lora !2025-03-28188641 | Instruct-tune LLaMA on consumer hardware| | 301|VinciGit00/Scrapegraph-ai !2025-03-281884618|Python scraper based on AI| | 302|gitroomhq/postiz-app !2025-03-281879110|📨 Schedule social posts, measure them, exchange with other members and get a lot of help from AI 🚀| | 303|PrefectHQ/prefect !2025-03-281878715|Prefect is a workflow orchestration tool empowering developers to build, observe, and react to data pipelines| | 304|ymcui/Chinese-LLaMA-Alpaca !2025-03-28187723 |Chinese LLaMA & Alpaca LLMs| | 305|kenjihiranabe/The-Art-of-Linear-Algebra !2025-03-28187335|Graphic notes on Gilbert Strang's "Linear Algebra for Everyone"| | 306|joonspk-research/generativeagents !2025-03-28187288|Generative Agents: Interactive Simulacra of Human Behavior| | 307|renovatebot/renovate !2025-03-28186820|Universal dependency update tool that fits into your workflows.| | 308|gventuri/pandas-ai !2025-03-28186109 |Pandas AI is a Python library that integrates generative artificial intelligence capabilities into Pandas, making dataframes conversational| | 309|thingsboard/thingsboard !2025-03-28185184|Open-source IoT Platform - Device management, data collection, processing and visualization.| | 310|ente-io/ente !2025-03-28184722|Fully open source, End to End Encrypted alternative to Google Photos and Apple Photos| | 311|serengil/deepface !2025-03-281840113|A Lightweight Face Recognition and Facial Attribute Analysis (Age, Gender, Emotion and Race) Library for Python| | 312|Raphire/Win11Debloat !2025-03-281840132|A simple, easy to use PowerShell script to remove pre-installed apps from windows, disable telemetry, remove Bing from windows search as well as perform various other changes to declutter and improve your windows experience. This script works for both windows 10 and windows 11.| | 313|Avaiga/taipy !2025-03-28179235|Turns Data and AI algorithms into production-ready web applications in no time.| | 314|microsoft/qlib !2025-03-281784231|Qlib is an AI-oriented quantitative investment platform that aims to realize the potential, empower research, and create value using AI technologies in quantitative investment, from exploring ideas to implementing productions. Qlib supports diverse machine learning modeling paradigms. including supervised learning, market dynamics modeling, and RL.| | 315|CopilotKit/CopilotKit !2025-03-281778571|Build in-app AI chatbots 🤖, and AI-powered Textareas ✨, into react web apps.| | 316|QwenLM/Qwen-7B !2025-03-281766017|The official repo of Qwen-7B (通义千问-7B) chat & pretrained large language model proposed by Alibaba Cloud.| | 317|w-okada/voice-changer !2025-03-28176078 |リアルタイムボイスチェンジャー Realtime Voice Changer| | 318|rlabbe/Kalman-and-Bayesian-Filters-in-Python !2025-03-281756011|Kalman Filter book using Jupyter Notebook. Focuses on building intuition and experience, not formal proofs. Includes Kalman filters,extended Kalman filters, unscented Kalman filters, particle filters, and more. All exercises include solutions.| | 319|Mikubill/sd-webui-controlnet !2025-03-28174794 |WebUI extension for ControlNet| | 320|jingyaogong/minimind !2025-03-2817380116|「大模型」3小时完全从0训练26M的小参数GPT，个人显卡即可推理训练！| | 321|apify/crawlee !2025-03-28172696|Crawlee—A web scraping and browser automation library for Node.js to build reliable crawlers. In JavaScript and TypeScript. Extract data for AI, LLMs, RAG, or GPTs. Download HTML, PDF, JPG, PNG, and other files from websites. Works with Puppeteer, Playwright, Cheerio, JSDOM, and raw HTTP. Both headful and headless mode. With proxy rotation.| | 322|apple/ml-stable-diffusion !2025-03-28172395|Stable Diffusion with Core ML on Apple Silicon| | 323| transitive-bullshit/chatgpt-api !2025-03-28172095 | Node.js client for the official ChatGPT API. | | 324|teableio/teable !2025-03-281719222|✨ The Next Gen Airtable Alternative: No-Code Postgres| | 325| xx025/carrot !2025-03-28170900 | Free ChatGPT Site List | | 326|microsoft/LightGBM !2025-03-28170723|A fast, distributed, high-performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.| | 327|VikParuchuri/surya !2025-03-28169827|Accurate line-level text detection and recognition (OCR) in any language| | 328|deepseek-ai/Janus !2025-03-281692825|Janus-Series: Unified Multimodal Understanding and Generation Models| | 329|ardalis/CleanArchitecture !2025-03-28168823|Clean Architecture Solution Template: A starting point for Clean Architecture with ASP.NET Core| | 330|neondatabase/neon !2025-03-28166466|Neon: Serverless Postgres. We separated storage and compute to offer autoscaling, code-like database branching, and scale to zero.| | 331|kestra-io/kestra !2025-03-281661313|⚡ Workflow Automation Platform. Orchestrate & Schedule code in any language, run anywhere, 500+ plugins. Alternative to Zapier, Rundeck, Camunda, Airflow...| | 332|Dao-AILab/flash-attention !2025-03-281659720|Fast and memory-efficient exact attention| | 333|RPCS3/rpcs3 !2025-03-281655712|PS3 emulator/debugger| | 334|meta-llama/llama-recipes !2025-03-28165486|Scripts for fine-tuning Llama2 with composable FSDP & PEFT methods to cover single/multi-node GPUs. Supports default & custom datasets for applications such as summarization & question answering. Supporting a number of candid inference solutions such as HF TGI, VLLM for local or cloud deployment.Demo apps to showcase Llama2 for WhatsApp & Messenger| | 335|emilwallner/Screenshot-to-code !2025-03-28165180|A neural network that transforms a design mock-up into a static website.| | 336|datawhalechina/llm-cookbook !2025-03-281650922|面向开发者的 LLM 入门教程，吴恩达大模型系列课程中文版| | 337|e2b-dev/awesome-ai-agents !2025-03-281643923|A list of AI autonomous agents| | 338|QwenLM/Qwen2.5 !2025-03-281641114|Qwen2.5 is the large language model series developed by Qwen team, Alibaba Cloud.| | 339|dair-ai/ML-YouTube-Courses !2025-03-28164114|📺 Discover the latest machine learning / AI courses on YouTube.| | 340|pybind/pybind11 !2025-03-28163620|Seamless operability between C++11 and Python| | 341|graphdeco-inria/gaussian-splatting !2025-03-281627116|Original reference implementation of "3D Gaussian Splatting for Real-Time Radiance Field Rendering"| | 342|meta-llama/codellama !2025-03-28162531|Inference code for CodeLlama models| | 343|TransformerOptimus/SuperAGI !2025-03-28161292 | SuperAGI - A dev-first open source autonomous AI agent framework. Enabling developers to build, manage & run useful autonomous agents quickly and reliably.| | 344|microsoft/onnxruntime !2025-03-28161169|ONNX Runtime: cross-platform, high-performance ML inferencing and training accelerator| | 345|IDEA-Research/Grounded-Segment-Anything !2025-03-281601411 |Marrying Grounding DINO with Segment Anything & Stable Diffusion & BLIP - Automatically Detect, Segment and Generate Anything with Image and Text Inputs| | 346|ddbourgin/numpy-ml !2025-03-28160054|Machine learning, in numpy| | 347|eosphoros-ai/DB-GPT !2025-03-281585225|Revolutionizing Database Interactions with Private LLM Technology| | 348|Stability-AI/StableLM !2025-03-28158310 |Stability AI Language Models| | 349|openai/evals !2025-03-28157935 |Evals is a framework for evaluating LLMs and LLM systems, and an open-source registry of benchmarks.| | 350|THUDM/ChatGLM2-6B !2025-03-28157500|ChatGLM2-6B: An Open Bilingual Chat LLM | | 351|sunner/ChatALL !2025-03-28156761 |Concurrently chat with ChatGPT, Bing Chat, Bard, Alpaca, Vincuna, Claude, ChatGLM, MOSS, iFlytek Spark, ERNIE and more, discover the best answers| | 352|abseil/abseil-cpp !2025-03-28156656|Abseil Common Libraries (C++)| | 353|NVIDIA/open-gpu-kernel-modules !2025-03-28156531|NVIDIA Linux open GPU kernel module source| | 354|letta-ai/letta !2025-03-281563718|Letta (formerly MemGPT) is a framework for creating LLM services with memory.| | 355|typescript-eslint/typescript-eslint !2025-03-28156211|✨ Monorepo for all the tooling which enables ESLint to support TypeScript| | 356|umijs/umi !2025-03-28156211|A framework in react community ✨| | 357|AI4Finance-Foundation/FinGPT !2025-03-281561215|Data-Centric FinGPT. Open-source for open finance! Revolutionize 🔥 We'll soon release the trained model.| | 358|amplication/amplication !2025-03-28156022|🔥🔥🔥 The Only Production-Ready AI-Powered Backend Code Generation| | 359|KindXiaoming/pykan !2025-03-28155477|Kolmogorov Arnold Networks| | 360|arc53/DocsGPT !2025-03-28154900|GPT-powered chat for documentation, chat with your documents| | 361|influxdata/telegraf !2025-03-28154502|Agent for collecting, processing, aggregating, and writing metrics, logs, and other arbitrary data.| | 362|microsoft/Bringing-Old-Photos-Back-to-Life !2025-03-28154084|Bringing Old Photo Back to Life (CVPR 2020 oral)| | 363|GaiZhenbiao/ChuanhuChatGPT !2025-03-2815394-2|GUI for ChatGPT API and many LLMs. Supports agents, file-based QA, GPT finetuning and query with web search. All with a neat UI.| | 364|Zeyi-Lin/HivisionIDPhotos !2025-03-281529710|⚡️HivisionIDPhotos: a lightweight and efficient AI ID photos tools. 一个轻量级的AI证件照制作算法。| | 365| mayooear/gpt4-pdf-chatbot-langchain !2025-03-281529518 | GPT4 & LangChain Chatbot for large PDF docs | | 366|1Panel-dev/MaxKB !2025-03-2815277148|? Based on LLM large language model knowledge base Q&A system. Ready to use out of the box, supports quick integration into third-party business systems. Officially produced by 1Panel| | 367|ai16z/eliza !2025-03-281526811|Conversational Agent for Twitter and Discord| | 368|apache/arrow !2025-03-28151684|Apache Arrow is a multi-language toolbox for accelerated data interchange and in-memory processing| | 369|princeton-nlp/SWE-agent !2025-03-281516119|SWE-agent: Agent Computer Interfaces Enable Software Engineering Language Models| | 370|mlc-ai/web-llm !2025-03-281509311 |Bringing large-language models and chat to web browsers. Everything runs inside the browser with no server support.| | 371|guillaumekln/faster-whisper !2025-03-281507117 |Faster Whisper transcription with CTranslate2| | 372|overleaf/overleaf !2025-03-28150316|A web-based collaborative LaTeX editor| | 373|triton-lang/triton !2025-03-28150169|Development repository for the Triton language and compiler| | 374|soxoj/maigret !2025-03-281500410|🕵️‍♂️ Collect a dossier on a person by username from thousands of sites| | 375|alibaba/lowcode-engine !2025-03-28149841|An enterprise-class low-code technology stack with scale-out design / 一套面向扩展设计的企业级低代码技术体系| | 376|espressif/esp-idf !2025-03-28148545|Espressif IoT Development Framework. Official development framework for Espressif SoCs.| | 377|pgvector/pgvector !2025-03-281484913|Open-source vector similarity search for Postgres| | 378|datawhalechina/leedl-tutorial !2025-03-28148246|《李宏毅深度学习教程》（李宏毅老师推荐👍），PDF下载地址：https://github.com/datawhalechina/leedl-tutorial/releases| | 379|xcanwin/KeepChatGPT !2025-03-28147972 |Using ChatGPT is more efficient and smoother, perfectly solving ChatGPT network errors. No longer do you need to frequently refresh the webpage, saving over 10 unnecessary steps| | 380|m-bain/whisperX !2025-03-281471313|WhisperX: Automatic Speech Recognition with Word-level Timestamps (& Diarization)| | 381|HumanAIGC/AnimateAnyone !2025-03-2814706-1|Animate Anyone: Consistent and Controllable Image-to-Video Synthesis for Character Animation| |!green-up-arrow.svg 382|naklecha/llama3-from-scratch !2025-03-281469024|llama3 implementation one matrix multiplication at a time| |!red-down-arrow 383| fauxpilot/fauxpilot !2025-03-28146871 | An open-source GitHub Copilot server | | 384|LlamaFamily/Llama-Chinese !2025-03-28145111|Llama Chinese Community, the best Chinese Llama large model, fully open source and commercially available| | 385|BradyFU/Awesome-Multimodal-Large-Language-Models !2025-03-281450121|Latest Papers and Datasets on Multimodal Large Language Models| | 386|vanna-ai/vanna !2025-03-281449819|🤖 Chat with your SQL database 📊. Accurate Text-to-SQL Generation via LLMs using RAG 🔄.| | 387|bleedline/aimoneyhunter !2025-03-28144845|AI Side Hustle Money Mega Collection: Teaching You How to Utilize AI for Various Side Projects to Earn Extra Income.| | 388|stefan-jansen/machine-learning-for-trading !2025-03-28144629|Code for Machine Learning for Algorithmic Trading, 2nd edition.| | 389|state-spaces/mamba !2025-03-28144139|Mamba: Linear-Time Sequence Modeling with Selective State Spaces| | 390|vercel/ai-chatbot !2025-03-281434614|A full-featured, hackable Next.js AI chatbot built by Vercel| | 391|steven-tey/novel !2025-03-281428410|Notion-style WYSIWYG editor with AI-powered autocompletions| | 392|unifyai/ivy !2025-03-281409348|Unified AI| | 393|chidiwilliams/buzz !2025-03-281402411 |Buzz transcribes and translates audio offline on your personal computer. Powered by OpenAI's Whisper.| | 394|lukas-blecher/LaTeX-OCR !2025-03-28139769|pix2tex: Using a ViT to convert images of equations into LaTeX code.| | 395|openai/tiktoken !2025-03-28139599|tiktoken is a fast BPE tokeniser for use with OpenAI's models.| | 396|nocobase/nocobase !2025-03-281391522|NocoBase is a scalability-first, open-source no-code/low-code platform for building business applications and enterprise solutions.| | 397|neonbjb/tortoise-tts !2025-03-28139010 |A multi-voice TTS system trained with an emphasis on quality| | 398|yamadashy/repomix !2025-03-281382036|📦 Repomix (formerly Repopack) is a powerful tool that packs your entire repository into a single, AI-friendly file. Perfect for when you need to feed your codebase to Large Language Models (LLMs) or other AI tools like Claude, ChatGPT, and Gemini.| | 399|adobe/react-spectrum !2025-03-28136766|A collection of libraries and tools that help you build adaptive, accessible, and robust user experiences.| | 400|THUDM/ChatGLM3 !2025-03-28136684|ChatGLM3 series: Open Bilingual Chat LLMs | | 401|NVIDIA/NeMo !2025-03-28134837|A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)| | 402|BlinkDL/RWKV-LM !2025-03-28134346 |RWKV is an RNN with transformer-level LLM performance. It can be directly trained like a GPT (parallelizable). So it combines the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding.| | 403| fuergaosi233/wechat-chatgpt !2025-03-28133330 | Use ChatGPT On Wechat via wechaty | | 404|udecode/plate !2025-03-28133325|A rich-text editor powered by AI| | 405|xenova/transformers.js !2025-03-281331219|State-of-the-art Machine Learning for the web. Run 🤗 Transformers directly in your browser, with no need for a server!| | 406|stas00/ml-engineering !2025-03-281325615|Machine Learning Engineering Guides and Tools| | 407| wong2/chatgpt-google-extension !2025-03-2813241-1 | A browser extension that enhances search engines with ChatGPT, this repos will not be updated from 2023-02-20| | 408|mrdbourke/pytorch-deep-learning !2025-03-281317520|Materials for the Learn PyTorch for Deep Learning: Zero to Mastery course.| | 409|Koenkk/zigbee2mqtt !2025-03-28131544|Zigbee 🐝 to MQTT bridge 🌉, get rid of your proprietary Zigbee bridges 🔨| | 410|vercel-labs/ai !2025-03-281298528|Build AI-powered applications with React, Svelte, and Vue| | 411|netease-youdao/QAnything !2025-03-28129318|Question and Answer based on Anything.| | 412|huggingface/trl !2025-03-281289622|Train transformer language models with reinforcement learning.| | 413|microsoft/BitNet !2025-03-28128503|Official inference framework for 1-bit LLMs| | 414|mediar-ai/screenpipe !2025-03-281283915|24/7 local AI screen & mic recording. Build AI apps that have the full context. Works with Ollama. Alternative to Rewind.ai. Open. Secure. You own your data. Rust.| | 415|Skyvern-AI/skyvern !2025-03-281277612|Automate browser-based workflows with LLMs and Computer Vision| | 416|pytube/pytube !2025-03-28126591|A lightweight, dependency-free Python library (and command-line utility) for downloading YouTube Videos.| | 417|official-stockfish/Stockfish !2025-03-28126574|UCI chess engine| | 418|sgl-project/sglang !2025-03-281260143|SGLang is a structured generation language designed for large language models (LLMs). It makes your interaction with LLMs faster and more controllable.| | 419|plasma-umass/scalene !2025-03-28125535|Scalene: a high-performance, high-precision CPU, GPU, and memory profiler for Python with AI-powered optimization proposals| | 420|danswer-ai/danswer !2025-03-28125503|Ask Questions in natural language and get Answers backed by private sources. Connects to tools like Slack, GitHub, Confluence, etc.| | 421|OpenTalker/SadTalker !2025-03-28125226|[CVPR 2023] SadTalker：Learning Realistic 3D Motion Coefficients for Stylized Audio-Driven Single Image Talking Face Animation| | 422|facebookresearch/AnimatedDrawings !2025-03-28123693 |Code to accompany "A Method for Animating Children's Drawings of the Human Figure"| | 423|activepieces/activepieces !2025-03-28123609|Your friendliest open source all-in-one automation tool ✨ Workflow automation tool 100+ integration / Enterprise automation tool / Zapier Alternative| | 424|ggerganov/ggml !2025-03-28121992 |Tensor library for machine learning| | 425|bytebase/bytebase !2025-03-28121694|World's most advanced database DevOps and CI/CD for Developer, DBA and Platform Engineering teams. The GitLab/GitHub for database DevOps.| | 426| willwulfken/MidJourney-Styles-and-Keywords-Reference !2025-03-28120971 | A reference containing Styles and Keywords that you can use with MidJourney AI| | 427|Huanshere/VideoLingo !2025-03-281207013|Netflix-level subtitle cutting, translation, alignment, and even dubbing - one-click fully automated AI video subtitle team | | 428|OpenLMLab/MOSS !2025-03-28120330 |An open-source tool-augmented conversational language model from Fudan University| | 429|llmware-ai/llmware !2025-03-281200727|Providing enterprise-grade LLM-based development framework, tools, and fine-tuned models.| | 430|PKU-YuanGroup/Open-Sora-Plan !2025-03-28119362|This project aim to reproduce Sora (Open AI T2V model), but we only have limited resource. We deeply wish the all open source community can contribute to this project.| | 431|ShishirPatil/gorilla !2025-03-28119332 |Gorilla: An API store for LLMs| | 432|NVIDIA/Megatron-LM !2025-03-281192716|Ongoing research training transformer models at scale| | 433|illacloud/illa-builder !2025-03-28119192|Create AI-Driven Apps like Assembling Blocks| | 434|marimo-team/marimo !2025-03-281191521|A reactive notebook for Python — run reproducible experiments, execute as a script, deploy as an app, and version with git.| | 435|smol-ai/developer !2025-03-28119111 | With 100k context windows on the way, it's now feasible for every dev to have their own smol developer| | 436|Lightning-AI/litgpt !2025-03-28118878|Pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.| | 437|openai/shap-e !2025-03-28118474 |Generate 3D objects conditioned on text or images| | 438|eugeneyan/open-llms !2025-03-28118451 |A list of open LLMs available for commercial use.| | 439|andrewyng/aisuite !2025-03-28118124|Simple, unified interface to multiple Generative AI providers| | 440|hajimehoshi/ebiten !2025-03-28117816|Ebitengine - A dead simple 2D game engine for Go| | 441|kgrzybek/modular-monolith-with-ddd !2025-03-28117493|Full Modular Monolith application with Domain-Driven Design approach.| | 442|h2oai/h2ogpt !2025-03-2811736-1 |Come join the movement to make the world's best open source GPT led by H2O.ai - 100% private chat and document search, no data leaks, Apache 2.0| | 443|owainlewis/awesome-artificial-intelligence !2025-03-28117332|A curated list of Artificial Intelligence (AI) courses, books, video lectures and papers.| | 444|DataTalksClub/mlops-zoomcamp !2025-03-28116643|Free MLOps course from DataTalks.Club| | 445|Rudrabha/Wav2Lip !2025-03-281163410|This repository contains the codes of "A Lip Sync Expert Is All You Need for Speech to Lip Generation In the Wild", published at ACM Multimedia 2020.| | 446|aishwaryanr/awesome-generative-ai-guide !2025-03-281152810|A one stop repository for generative AI research updates, interview resources, notebooks and much more!| | 447|karpathy/micrograd !2025-03-28115146|A tiny scalar-valued autograd engine and a neural net library on top of it with PyTorch-like API| | 448|InstantID/InstantID !2025-03-28115111|InstantID : Zero-shot Identity-Preserving Generation in Seconds 🔥| | 449|facebookresearch/seamlesscommunication !2025-03-28114434|Foundational Models for State-of-the-Art Speech and Text Translation| | 450|anthropics/anthropic-cookbook !2025-03-281140112|A collection of notebooks/recipes showcasing some fun and effective ways of using Claude.| | 451|mastra-ai/mastra !2025-03-281139240|the TypeScript AI agent framework| | 452|NVIDIA/TensorRT !2025-03-28113864|NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.| | 453|plandex-ai/plandex !2025-03-28113645|An AI coding engine for complex tasks| | 454|RUCAIBox/LLMSurvey !2025-03-28112735 |A collection of papers and resources related to Large Language Models.| | 455|kubeshark/kubeshark !2025-03-28112711|The API traffic analyzer for Kubernetes providing real-time K8s protocol-level visibility, capturing and monitoring all traffic and payloads going in, out and across containers, pods, nodes and clusters. Inspired by Wireshark, purposely built for Kubernetes| | 456|electric-sql/pglite !2025-03-28112617|Lightweight Postgres packaged as WASM into a TypeScript library for the browser, Node.js, Bun and Deno from https://electric-sql.com| | 457|lightaime/camel !2025-03-281124441 |🐫 CAMEL: Communicative Agents for “Mind” Exploration of Large Scale Language Model Society| | 458|huggingface/lerobot !2025-03-281120184|🤗 LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch| | 459|normal-computing/outlines !2025-03-28111657|Generative Model Programming| | 460|libretro/RetroArch !2025-03-28110701|Cross-platform, sophisticated frontend for the libretro API. Licensed GPLv3.| | 461|THUDM/CogVideo !2025-03-28110599|Text-to-video generation: CogVideoX (2024) and CogVideo (ICLR 2023)| | 462|bentoml/OpenLLM !2025-03-28110495|An open platform for operating large language models (LLMs) in production. Fine-tune, serve, deploy, and monitor any LLMs with ease.| | 463|vosen/ZLUDA !2025-03-28110429|CUDA on AMD GPUs| | 464|dair-ai/ML-Papers-of-the-Week !2025-03-28110304 |🔥Highlighting the top ML papers every week.| | 465|WordPress/gutenberg !2025-03-28110212|The Block Editor project for WordPress and beyond. Plugin is available from the official repository.| | 466|microsoft/data-formulator !2025-03-281099827|🪄 Create rich visualizations with AI| | 467|LibreTranslate/LibreTranslate !2025-03-28109887|Free and Open Source Machine Translation API. Self-hosted, offline capable and easy to setup.| | 468|block/goose !2025-03-281097737|an open-source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM| | 469|getumbrel/llama-gpt !2025-03-28109553|A self-hosted, offline, ChatGPT-like chatbot. Powered by Llama 2. 100% private, with no data leaving your device.| | 470|HigherOrderCO/HVM !2025-03-28109182|A massively parallel, optimal functional runtime in Rust| | 471|databrickslabs/dolly !2025-03-2810812-3 | A large language model trained on the Databricks Machine Learning Platform| | 472|srush/GPU-Puzzles !2025-03-28108014|Solve puzzles. Learn CUDA.| | 473|Z3Prover/z3 !2025-03-28107952|The Z3 Theorem Prover| | 474|UFund-Me/Qbot !2025-03-281079313 |Qbot is an AI-oriented quantitative investment platform, which aims to realize the potential, empower AI technologies in quantitative investment| | 475|langchain-ai/langgraph !2025-03-281077336|| | 476|lz4/lz4 !2025-03-28107647|Extremely Fast Compression algorithm| | 477|magic-research/magic-animate !2025-03-28107160|MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model| | 478|PaperMC/Paper !2025-03-281071410|The most widely used, high performance Minecraft server that aims to fix gameplay and mechanics inconsistencies| | 479|getomni-ai/zerox !2025-03-281071015|Zero shot pdf OCR with gpt-4o-mini| |!green-up-arrow.svg 480|🔥NirDiamant/GenAIAgents !2025-03-2810693318|This repository provides tutorials and implementations for various Generative AI Agent techniques, from basic to advanced. It serves as a comprehensive guide for building intelligent, interactive AI systems.| |!red-down-arrow 481|Unstructured-IO/unstructured !2025-03-28106889|Open source libraries and APIs to build custom preprocessing pipelines for labeling, training, or production machine learning pipelines.| | 482|apache/thrift !2025-03-28106610|Apache Thrift| | 483| TheR1D/shellgpt !2025-03-28106097 | A command-line productivity tool powered by ChatGPT, will help you accomplish your tasks faster and more efficiently | | 484|TheRamU/Fay !2025-03-281060312 |Fay is a complete open source project that includes Fay controller and numeral models, which can be used in different applications such as virtual hosts, live promotion, numeral human interaction and so on| | 485|zyronon/douyin !2025-03-28105566|Vue3 + Pinia + Vite5 仿抖音，Vue 在移动端的最佳实践 . Imitate TikTok ，Vue Best practices on Mobile| | 486|THU-MIG/yolov10 !2025-03-28105485|YOLOv10: Real-Time End-to-End Object Detection| | 487|idootop/mi-gpt !2025-03-281052522|? Transform XiaoAi speaker into a personal voice assistant with ChatGPT and DouBao integration.| | 488|SakanaAI/AI-Scientist !2025-03-281051310|The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery 🧑‍🔬| | 489|szimek/sharedrop !2025-03-28105101|Easy P2P file transfer powered by WebRTC - inspired by Apple AirDrop| | 490|salesforce/LAVIS !2025-03-28103942 |LAVIS - A One-stop Library for Language-Vision Intelligence| | 491|aws/amazon-sagemaker-examples !2025-03-28103654|Example 📓 Jupyter notebooks that demonstrate how to build, train, and deploy machine learning models using 🧠 Amazon SageMaker.| | 492|artidoro/qlora !2025-03-28103402 |QLoRA: Efficient Finetuning of Quantized LLMs| | 493|lllyasviel/stable-diffusion-webui-forge !2025-03-281029314| a platform on top of Stable Diffusion WebUI (based on Gradio) to make development easier, optimize resource management, and speed up inference| | 494|NielsRogge/Transformers-Tutorials !2025-03-28102487|This repository contains demos I made with the Transformers library by HuggingFace.| | 495|kedro-org/kedro !2025-03-28102371|Kedro is a toolbox for production-ready data science. It uses software engineering best practices to help you create data engineering and data science pipelines that are reproducible, maintainable, and modular.| | 496| chathub-dev/chathub !2025-03-28102301 | All-in-one chatbot client | | 497|microsoft/promptflow !2025-03-28101612|Build high-quality LLM apps - from prototyping, testing to production deployment and monitoring.| | 498|mistralai/mistral-src !2025-03-28101372|Reference implementation of Mistral AI 7B v0.1 model.| | 499|burn-rs/burn !2025-03-28101183|Burn - A Flexible and Comprehensive Deep Learning Framework in Rust| | 500|AIGC-Audio/AudioGPT !2025-03-28101150 |AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head| | 501|facebookresearch/dinov2 !2025-03-281011210 |PyTorch code and models for the DINOv2 self-supervised learning method.| | 502|RockChinQ/LangBot !2025-03-281008455|😎丰富生态、🧩支持扩展、🦄多模态 - 大模型原生即时通信机器人平台 🤖 | | 503|78/xiaozhi-esp32 !2025-03-281008180|Build your own AI friend| | 504|cumulo-autumn/StreamDiffusion !2025-03-28100761|StreamDiffusion: A Pipeline-Level Solution for Real-Time Interactive Generation| | 505|DataTalksClub/machine-learning-zoomcamp !2025-03-28100664|The code from the Machine Learning Bookcamp book and a free course based on the book| | 506|nerfstudio-project/nerfstudio !2025-03-28100343|A collaboration friendly studio for NeRFs| | 507|cupy/cupy !2025-03-28100344|NumPy & SciPy for GPU| | 508|NVIDIA/TensorRT-LLM !2025-03-281000823|TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM also contains components to create Python and C++ runtimes that execute those TensorRT engines.| | 509|wasp-lang/open-saas !2025-03-2899665|A free, open-source SaaS app starter for React & Node.js with superpowers. Production-ready. Community-driven.| | 510|huggingface/text-generation-inference !2025-03-2899383|Large Language Model Text Generation Inference| | 511|jxnl/instructor !2025-03-2899224|structured outputs for llms| | 512|GoogleCloudPlatform/generative-ai !2025-03-2899086|Sample code and notebooks for Generative AI on Google Cloud| | 513|manticoresoftware/manticoresearch !2025-03-2898799|Easy to use open source fast database for search | | 514|langfuse/langfuse !2025-03-28985134|🪢 Open source LLM engineering platform. Observability, metrics, evals, prompt management, testing, prompt playground, datasets, LLM evaluations -- 🍊YC W23 🤖 integrate via Typescript, Python / Decorators, OpenAI, Langchain, LlamaIndex, Litellm, Instructor, Mistral, Perplexity, Claude, Gemini, Vertex| | 515|keephq/keep !2025-03-2897949|The open-source alert management and AIOps platform| | 516|sashabaranov/go-openai !2025-03-2897843|OpenAI ChatGPT, GPT-3, GPT-4, DALL·E, Whisper API wrapper for Go| | 517|autowarefoundation/autoware !2025-03-2897766|Autoware - the world's leading open-source software project for autonomous driving| | 518|anthropics/courses !2025-03-2897269|Anthropic's educational courses| | 519|popcorn-official/popcorn-desktop !2025-03-2896853|Popcorn Time is a multi-platform, free software BitTorrent client that includes an integrated media player ( Windows / Mac / Linux ) A Butter-Project Fork| | 520|getmaxun/maxun !2025-03-28968515|🔥 Open-source no-code web data extraction platform. Turn websites to APIs and spreadsheets with no-code robots in minutes! [In Beta]| | 521|wandb/wandb !2025-03-2896763|🔥 A tool for visualizing and tracking your machine learning experiments. This repo contains the CLI and Python API.| | 522|karpathy/minbpe !2025-03-2895353|Minimal, clean, code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization.| | 523|bigscience-workshop/petals !2025-03-2895142|🌸 Run large language models at home, BitTorrent-style. Fine-tuning and inference up to 10x faster than offloading| | 524|OthersideAI/self-operating-computer !2025-03-2894931|A framework to enable multimodal models to operate a computer.| | 525|mshumer/gpt-prompt-engineer !2025-03-2894911|| | 526| BloopAI/bloop !2025-03-2894710 | A fast code search engine written in Rust| | 527|BlinkDL/ChatRWKV !2025-03-289467-1 |ChatRWKV is like ChatGPT but powered by RWKV (100% RNN) language model, and open source.| | 528|timlrx/tailwind-nextjs-starter-blog !2025-03-2894677|This is a Next.js, Tailwind CSS blogging starter template. Comes out of the box configured with the latest technologies to make technical writing a breeze. Easily configurable and customizable. Perfect as a replacement to existing Jekyll and Hugo individual blogs.| | 529|google/benchmark !2025-03-2893634|A microbenchmark support library| | 530|facebookresearch/nougat !2025-03-2893603|Implementation of Nougat Neural Optical Understanding for Academic Documents| | 531|modelscope/facechain !2025-03-2893536|FaceChain is a deep-learning toolchain for generating your Digital-Twin.| | 532|DrewThomasson/ebook2audiobook !2025-03-2893388|Convert ebooks to audiobooks with chapters and metadata using dynamic AI models and voice cloning. Supports 1,107+ languages!| | 533|RayTracing/raytracing.github.io !2025-03-2893035|Main Web Site (Online Books)| | 534|QwenLM/Qwen2.5-VL !2025-03-28930249|Qwen2.5-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.| | 535|WongKinYiu/yolov9 !2025-03-2892201|Implementation of paper - YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information| | 536|alibaba-damo-academy/FunASR !2025-03-28920222|A Fundamental End-to-End Speech Recognition Toolkit and Open Source SOTA Pretrained Models.| | 537|Visualize-ML/Book4Power-of-Matrix !2025-03-2891931|Book4 'Power of Matrix' | | 538|dice2o/BingGPT !2025-03-289185-1 |Desktop application of new Bing's AI-powered chat (Windows, macOS and Linux)| | 539|browserbase/stagehand !2025-03-28917621|An AI web browsing framework focused on simplicity and extensibility.| | 540|FlagOpen/FlagEmbedding !2025-03-28914111|Dense Retrieval and Retrieval-augmented LLMs| | 541|Const-me/Whisper !2025-03-2890979|High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model| | 542|lucidrains/denoising-diffusion-pytorch !2025-03-2890942|Implementation of Denoising Diffusion Probabilistic Model in Pytorch| | 543|Chainlit/chainlit !2025-03-28904422|Build Conversational AI in minutes ⚡️| | 544|togethercomputer/OpenChatKit !2025-03-2890160 |OpenChatKit provides a powerful, open-source base to create both specialized and general purpose chatbots for various applications| | 545|Stability-AI/StableStudio !2025-03-2889631 |Community interface for generative AI| | 546|voicepaw/so-vits-svc-fork !2025-03-2889482 |so-vits-svc fork with realtime support, improved interface and more features.| | 547|pymc-devs/pymc !2025-03-2889413|Bayesian Modeling and Probabilistic Programming in Python| | 548|espnet/espnet !2025-03-2889302|End-to-End Speech Processing Toolkit| | 549|kedacore/keda !2025-03-2888991|KEDA is a Kubernetes-based Event Driven Autoscaling component. It provides event driven scale for any container running in Kubernetes| | 550|open-mmlab/Amphion !2025-03-28886911|Amphion (/æmˈfaɪən/) is a toolkit for Audio, Music, and Speech Generation. Its purpose is to support reproducible research and help junior researchers and engineers get started in the field of audio, music, and speech generation research and development.| | 551|gorse-io/gorse !2025-03-2888451|Gorse open source recommender system engine| | 552|adams549659584/go-proxy-bingai !2025-03-288768-1 |A Microsoft New Bing demo site built with Vue3 and Go, providing a consistent UI experience, supporting ChatGPT prompts, and accessible within China| | 553|open-mmlab/mmsegmentation !2025-03-2887513|OpenMMLab Semantic Segmentation Toolbox and Benchmark.| | 554|bytedance/monolith !2025-03-2887223|ByteDance's Recommendation System| | 555|LouisShark/chatgptsystemprompt !2025-03-2887216|store all agent's system prompt| | 556|brexhq/prompt-engineering !2025-03-2887080 |Tips and tricks for working with Large Language Models like OpenAI's GPT-4.| | 557|erincatto/box2d !2025-03-2886841|Box2D is a 2D physics engine for games| | 558|🔥microsoft/ai-agents-for-beginners !2025-03-288669323|10 Lessons to Get Started Building AI Agents| | 559|nashsu/FreeAskInternet !2025-03-2886102|FreeAskInternet is a completely free, private and locally running search aggregator & answer generate using LLM, without GPU needed. The user can ask a question and the system will make a multi engine search and combine the search result to the ChatGPT3.5 LLM and generate the answer based on search results.| | 560|goldmansachs/gs-quant !2025-03-2885981|Python toolkit for quantitative finance| | 561|srbhr/Resume-Matcher !2025-03-2885800|Open Source Free ATS Tool to compare Resumes with Job Descriptions and create a score to rank them.| | 562|facebookresearch/ImageBind !2025-03-2885681 |ImageBind One Embedding Space to Bind Them All| | 563|ashawkey/stable-dreamfusion !2025-03-2885481 |A pytorch implementation of text-to-3D dreamfusion, powered by stable diffusion.| | 564|meetecho/janus-gateway !2025-03-2885232|Janus WebRTC Server| | 565|google/magika !2025-03-2885003|Detect file content types with deep learning| | 566|huggingface/chat-ui !2025-03-2884871 |Open source codebase powering the HuggingChat app| | 567|EleutherAI/lm-evaluation-harness !2025-03-28843012|A framework for few-shot evaluation of autoregressive language models.| | 568|jina-ai/reader !2025-03-2884089|Convert any URL to an LLM-friendly input with a simple prefix https://r.jina.ai/| | 569|microsoft/TypeChat !2025-03-288406-1|TypeChat is a library that makes it easy to build natural language interfaces using types.| | 570|thuml/Time-Series-Library !2025-03-28839715|A Library for Advanced Deep Time Series Models.| | 571|OptimalScale/LMFlow !2025-03-2883882|An Extensible Toolkit for Finetuning and Inference of Large Foundation Models. Large Model for All.| | 572|baptisteArno/typebot.io !2025-03-2883845|💬 Typebot is a powerful chatbot builder that you can self-host.| | 573|jzhang38/TinyLlama !2025-03-2883504|The TinyLlama project is an open endeavor to pretrain a 1.1B Llama model on 3 trillion tokens.| | 574|fishaudio/Bert-VITS2 !2025-03-2883472|vits2 backbone with multilingual-bert| | 575|OpenBMB/XAgent !2025-03-2882683|An Autonomous LLM Agent for Complex Task Solving| | 576|Acly/krita-ai-diffusion !2025-03-2882387|Streamlined interface for generating images with AI in Krita. Inpaint and outpaint with optional text prompt, no tweaking required.| | 577|jasonppy/VoiceCraft !2025-03-2882151|Zero-Shot Speech Editing and Text-to-Speech in the Wild| | 578|SJTU-IPADS/PowerInfer !2025-03-2881693|High-speed Large Language Model Serving on PCs with Consumer-grade GPUs| | 579|modelscope/DiffSynth-Studio !2025-03-28814713|Enjoy the magic of Diffusion models!| | 580|o3de/o3de !2025-03-2881443|Open 3D Engine (O3DE) is an Apache 2.0-licensed multi-platform 3D engine that enables developers and content creators to build AAA games, cinema-quality 3D worlds, and high-fidelity simulations without any fees or commercial obligations.| | 581|zmh-program/chatnio !2025-03-2881325|🚀 Next Generation AI One-Stop Internationalization Solution. 🚀 下一代 AI 一站式 B/C 端解决方案，支持 OpenAI，Midjourney，Claude，讯飞星火，Stable Diffusion，DALL·E，ChatGLM，通义千问，腾讯混元，360 智脑，百川 AI，火山方舟，新必应，Gemini，Moonshot 等模型，支持对话分享，自定义预设，云端同步，模型市场，支持弹性计费和订阅计划模式，支持图片解析，支持联网搜索，支持模型缓存，丰富美观的后台管理与仪表盘数据统计。| | 582|leptonai/searchwithlepton !2025-03-2880632|Building a quick conversation-based search demo with Lepton AI.| | 583|sebastianstarke/AI4Animation !2025-03-2880620|Bringing Characters to Life with Computer Brains in Unity| | 584|wangrongding/wechat-bot !2025-03-2880528|🤖一个基于 WeChaty 结合 DeepSeek / ChatGPT / Kimi / 讯飞等Ai服务实现的微信机器人，可以用来帮助你自动回复微信消息，或者管理微信群/好友，检测僵尸粉等...| | 585|openvinotoolkit/openvino !2025-03-2880528|OpenVINO™ is an open-source toolkit for optimizing and deploying AI inference| | 586|steven2358/awesome-generative-ai !2025-03-28802610|A curated list of modern Generative Artificial Intelligence projects and services| | 587|adam-maj/tiny-gpu !2025-03-2880234|A minimal GPU design in Verilog to learn how GPUs work from the ground up| | 588| anse-app/chatgpt-demo !2025-03-2880180 | A demo repo based on OpenAI API (gpt-3.5-turbo) | | 589| acheong08/EdgeGPT !2025-03-288015-1 |Reverse engineered API of Microsoft's Bing Chat | | 590|ai-collection/ai-collection !2025-03-2879994 |The Generative AI Landscape - A Collection of Awesome Generative AI Applications| | 591|GreyDGL/PentestGPT !2025-03-2879953 |A GPT-empowered penetration testing tool| | 592|delta-io/delta !2025-03-2879112|An open-source storage framework that enables building a Lakehouse architecture with compute engines including Spark, PrestoDB, Flink, Trino, and Hive and APIs| | 593|dataelement/bisheng !2025-03-2879085|Bisheng is an open LLM devops platform for next generation AI applications.| | 594|e2b-dev/e2b !2025-03-2878447 |Vercel for AI agents. We help developers to build, deploy, and monitor AI agents. Focusing on specialized AI agents that build software for you - your personal software developers.| | 595|01-ai/Yi !2025-03-2878311|A series of large language models trained from scratch by developers @01-ai| | 596|Plachtaa/VALL-E-X !2025-03-287830-1|An open source implementation of Microsoft's VALL-E X zero-shot TTS model. The demo is available at https://plachtaa.github.io| | 597|abhishekkrthakur/approachingalmost !2025-03-2878204|Approaching (Almost) Any Machine Learning Problem| | 598|pydantic/pydantic-ai !2025-03-28781041|Agent Framework / shim to use Pydantic with LLMs| | 599|rany2/edge-tts !2025-03-2877901|Use Microsoft Edge's online text-to-speech service from Python WITHOUT needing Microsoft Edge or Windows or an API key| | 600|CASIA-IVA-Lab/FastSAM !2025-03-2877881|Fast Segment Anything| | 601|netease-youdao/EmotiVoice !2025-03-2877817|EmotiVoice 😊: a Multi-Voice and Prompt-Controlled TTS Engine| | 602|lllyasviel/IC-Light !2025-03-2877804|More relighting!| | 603|kroma-network/tachyon !2025-03-287774-1|Modular ZK(Zero Knowledge) backend accelerated by GPU| | 604|deep-floyd/IF !2025-03-2877731 |A novel state-of-the-art open-source text-to-image model with a high degree of photorealism and language understanding| | 605|oumi-ai/oumi !2025-03-2877705|Everything you need to build state-of-the-art foundation models, end-to-end.| | 606|reorproject/reor !2025-03-2877681|AI note-taking app that runs models locally.| | 607|lightpanda-io/browser !2025-03-28775813|Lightpanda: the headless browser designed for AI and automation| | 608|xiangsx/gpt4free-ts !2025-03-287755-1|Providing a free OpenAI GPT-4 API ! This is a replication project for the typescript version of xtekky/gpt4free| | 609|IDEA-Research/GroundingDINO !2025-03-28773311|Official implementation of the paper "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection"| | 610|bunkerity/bunkerweb !2025-03-2877326|🛡️ Make your web services secure by default !| | 611|vikhyat/moondream !2025-03-2877057|tiny vision language model| | 612|firmai/financial-machine-learning !2025-03-287703-1|A curated list of practical financial machine learning tools and applications.| | 613|n8n-io/self-hosted-ai-starter-kit !2025-03-28765121|The Self-hosted AI Starter Kit is an open-source template that quickly sets up a local AI environment. Curated by n8n, it provides essential tools for creating secure, self-hosted AI workflows.| | 614|intel-analytics/ipex-llm !2025-03-2876507|Accelerate local LLM inference and finetuning (LLaMA, Mistral, ChatGLM, Qwen, Baichuan, Mixtral, Gemma, etc.) on Intel CPU and GPU (e.g., local PC with iGPU, discrete GPU such as Arc, Flex and Max). A PyTorch LLM library that seamlessly integrates with llama.cpp, HuggingFace, LangChain, LlamaIndex, DeepSpeed, vLLM, FastChat, ModelScope, etc.| | 615|jrouwe/JoltPhysics !2025-03-28764510|A multi core friendly rigid body physics and collision detection library. Written in C++. Suitable for games and VR applications. Used by Horizon Forbidden West.| | 616|THUDM/CodeGeeX2 !2025-03-2876270|CodeGeeX2: A More Powerful Multilingual Code Generation Model| | 617|meta-llama/llama-stack !2025-03-2875866|Composable building blocks to build Llama Apps| | 618|sweepai/sweep !2025-03-287530-1|Sweep is an AI junior developer| | 619|lllyasviel/Omost !2025-03-2875301|Your image is almost there!| | 620|ahmedbahaaeldin/From-0-to-Research-Scientist-resources-guide !2025-03-2875050|Detailed and tailored guide for undergraduate students or anybody want to dig deep into the field of AI with solid foundation.| | 621|dair-ai/ML-Papers-Explained !2025-03-2875050|Explanation to key concepts in ML| | 622|zaidmukaddam/scira !2025-03-28750110|Scira (Formerly MiniPerplx) is a minimalistic AI-powered search engine that helps you find information on the internet. Powered by Vercel AI SDK! Search with models like Grok 2.0.| | 623|Portkey-AI/gateway !2025-03-28749416|A Blazing Fast AI Gateway. Route to 100+ LLMs with 1 fast & friendly API.| | 624|web-infra-dev/midscene !2025-03-28748729|An AI-powered automation SDK can control the page, perform assertions, and extract data in JSON format using natural language.| | 625|zilliztech/GPTCache !2025-03-2874801 |GPTCache is a library for creating semantic cache to store responses from LLM queries.| | 626|niedev/RTranslator !2025-03-2874742|RTranslator is the world's first open source real-time translation app.| |!green-up-arrow.svg 627|roboflow/notebooks !2025-03-2874666|Examples and tutorials on using SOTA computer vision models and techniques. Learn everything from old-school ResNet, through YOLO and object-detection transformers like DETR, to the latest models like Grounding DINO and SAM.| |!red-down-arrow 628|openlm-research/openllama !2025-03-2874652|OpenLLaMA, a permissively licensed open source reproduction of Meta AI’s LLaMA 7B trained on the RedPajama dataset| | 629|LiheYoung/Depth-Anything !2025-03-2874155|Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data| | 630|enso-org/enso !2025-03-2874040|Hybrid visual and textual functional programming.| | 631|bigcode-project/starcoder !2025-03-287401-1 |Home of StarCoder: fine-tuning & inference!| | 632|git-ecosystem/git-credential-manager !2025-03-2873975|Secure, cross-platform Git credential storage with authentication to GitHub, Azure Repos, and other popular Git hosting services.| | 633|OpenGVLab/InternVL !2025-03-2873634|[CVPR 2024 Oral] InternVL Family: A Pioneering Open-Source Alternative to GPT-4V. 接近GPT-4V表现的可商用开源模型| | 634|WooooDyy/LLM-Agent-Paper-List !2025-03-2873551|The paper list of the 86-page paper "The Rise and Potential of Large Language Model Based Agents: A Survey" by Zhiheng Xi et al.| | 635|lencx/Noi !2025-03-2873157|🦄 AI + Tools + Plugins + Community| | 636|udlbook/udlbook !2025-03-2873075|Understanding Deep Learning - Simon J.D. Prince| | 637|OpenBMB/MiniCPM !2025-03-2872841|MiniCPM-2B: An end-side LLM outperforms Llama2-13B.| | 638|jaywalnut310/vits !2025-03-2872815 |VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech| | 639|xorbitsai/inference !2025-03-28727528|Replace OpenAI GPT with another LLM in your app by changing a single line of code. Xinference gives you the freedom to use any LLM you need. With Xinference, you're empowered to run inference with any open-source language models, speech recognition models, and multimodal models, whether in the cloud, on-premises, or even on your laptop.| | 640|PWhiddy/PokemonRedExperiments !2025-03-2872492|Playing Pokemon Red with Reinforcement Learning| | 641|Canner/WrenAI !2025-03-28723213|🤖 Open-source AI Agent that empowers data-driven teams to chat with their data to generate Text-to-SQL, charts, spreadsheets, reports, and BI. 📈📊📋🧑‍💻| | 642|miurla/morphic !2025-03-2872258|An AI-powered answer engine with a generative UI| | 643|ml-explore/mlx-examples !2025-03-2872168|Examples in the MLX framework| | 644|PKU-YuanGroup/ChatLaw !2025-03-2872010|Chinese Legal Large Model| | 645|NVIDIA/cutlass !2025-03-2871883|CUDA Templates for Linear Algebra Subroutines| | 646|FoundationVision/VAR !2025-03-28717444|[GPT beats diffusion🔥] [scaling laws in visual generation📈] Official impl. of "Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction"| | 647|ymcui/Chinese-LLaMA-Alpaca-2 !2025-03-2871561|Chinese LLaMA-2 & Alpaca-2 LLMs| | 648|nadermx/backgroundremover !2025-03-2871514 |Background Remover lets you Remove Background from images and video using AI with a simple command line interface that is free and open source.| | 649|onuratakan/gpt-computer-assistant !2025-03-28714514|gpt-4o for windows, macos and ubuntu| | 650|graviraja/MLOps-Basics !2025-03-2871326|| | 651|Future-House/paper-qa !2025-03-287118-1|High accuracy RAG for answering questions from scientific documents with citations| | 652|open-mmlab/mmagic !2025-03-2871102 |OpenMMLab Multimodal Advanced, Generative, and Intelligent Creation Toolbox| | 653|bhaskatripathi/pdfGPT !2025-03-2870941 |PDF GPT allows you to chat with the contents of your PDF file by using GPT capabilities. The only open source solution to turn your pdf files in a chatbot!| | 654|ollama/ollama-python !2025-03-28709117|Ollama Python library| | 655|facebookresearch/DiT !2025-03-2870376|Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"| | 656|geekyutao/Inpaint-Anything !2025-03-2870262 |Inpaint anything using Segment Anything and inpainting models.| | 657|AbdullahAlfaraj/Auto-Photoshop-StableDiffusion-Plugin !2025-03-2870160 |A user-friendly plug-in that makes it easy to generate stable diffusion images inside Photoshop using Automatic1111-sd-webui as a backend.| | 658|apple/corenet !2025-03-2869990|CoreNet: A library for training deep neural networks| | 659|openstatusHQ/openstatus !2025-03-2869926|🏓 The open-source synthetic monitoring platform 🏓| | 660|weaviate/Verba !2025-03-2869772|Retrieval Augmented Generation (RAG) chatbot powered by Weaviate| | 661|meshery/meshery !2025-03-2869630|Meshery, the cloud native manager| | 662|OpenTalker/video-retalking !2025-03-2869530|[SIGGRAPH Asia 2022] VideoReTalking: Audio-based Lip Synchronization for Talking Head Video Editing In the Wild| | 663|digitalinnovationone/dio-lab-open-source !2025-03-28689013|Repositório do lab "Contribuindo em um Projeto Open Source no GitHub" da Digital Innovation One.| | 664|jianchang512/ChatTTS-ui !2025-03-2868842|一个简单的本地网页界面，直接使用ChatTTS将文字合成为语音，同时支持对外提供API接口。| | 665|patchy631/ai-engineering-hub !2025-03-28686434|In-depth tutorials on LLMs, RAGs and real-world AI agent applications.| | 666|gunnarmorling/1brc !2025-03-2868512|1️⃣🐝🏎️ The One Billion Row Challenge -- A fun exploration of how quickly 1B rows from a text file can be aggregated with Java| | 667|Azure-Samples/azure-search-openai-demo !2025-03-2868482 |A sample app for the Retrieval-Augmented Generation pattern running in Azure, using Azure Cognitive Search for retrieval and Azure OpenAI large language models to power ChatGPT-style and Q&A experiences.| | 668|mit-han-lab/streaming-llm !2025-03-2868382|Efficient Streaming Language Models with Attention Sinks| | 669|InternLM/InternLM !2025-03-2868352|InternLM has open-sourced a 7 billion parameter base model, a chat model tailored for practical scenarios and the training system.| | 670|dependency-check/DependencyCheck !2025-03-2868191|OWASP dependency-check is a software composition analysis utility that detects publicly disclosed vulnerabilities in application dependencies.| | 671|Soulter/AstrBot !2025-03-28678643|✨易上手的多平台 LLM 聊天机器人及开发框架✨。支持 QQ、QQ频道、Telegram、微信平台(Gewechat, 企业微信)、内置 Web Chat，OpenAI GPT、DeepSeek、Ollama、Llama、GLM、Gemini、OneAPI、LLMTuner，支持 LLM Agent 插件开发，可视化面板。一键部署。支持 Dify 工作流、代码执行器、Whisper 语音转文字。| | 672|react-native-webview/react-native-webview !2025-03-2867792|React Native Cross-Platform WebView| | 673|modelscope/agentscope !2025-03-28676916|Start building LLM-empowered multi-agent applications in an easier way.| | 674|mylxsw/aidea !2025-03-2867381|AIdea is a versatile app that supports GPT and domestic large language models，also supports "Stable Diffusion" text-to-image generation, image-to-image generation, SDXL 1.0, super-resolution, and image colorization| | 675|langchain-ai/ollama-deep-researcher !2025-03-28668635|Fully local web research and report writing assistant| | 676|threestudio-project/threestudio !2025-03-2866653|A unified framework for 3D content generation.| | 677|gaomingqi/Track-Anything !2025-03-2866631 |A flexible and interactive tool for video object tracking and segmentation, based on Segment Anything, XMem, and E2FGVI.| | 678|spdustin/ChatGPT-AutoExpert !2025-03-2866570|🚀🧠💬 Supercharged Custom Instructions for ChatGPT (non-coding) and ChatGPT Advanced Data Analysis (coding).| | 679|HariSekhon/DevOps-Bash-tools !2025-03-2866463|1000+ DevOps Bash Scripts - AWS, GCP, Kubernetes, Docker, CI/CD, APIs, SQL, PostgreSQL, MySQL, Hive, Impala, Kafka, Hadoop, Jenkins, GitHub, GitLab, BitBucket, Azure DevOps, TeamCity, Spotify, MP3, LDAP, Code/Build Linting, pkg mgmt for Linux, Mac, Python, Perl, Ruby, NodeJS, Golang, Advanced dotfiles: .bashrc, .vimrc, .gitconfig, .screenrc, tmux..| | 680|modelscope/swift !2025-03-28661530|ms-swift: Use PEFT or Full-parameter to finetune 200+ LLMs or 15+ MLLMs| | 681|langchain-ai/opengpts !2025-03-2866080|This is an open source effort to create a similar experience to OpenAI's GPTs and Assistants API| | 682| yihong0618/xiaogpt !2025-03-2865131 | Play ChatGPT with xiaomi ai speaker | | 683| civitai/civitai !2025-03-2865111 | Build a platform where people can share their stable diffusion models | | 684|KoljaB/RealtimeSTT !2025-03-28649513|A robust, efficient, low-latency speech-to-text library with advanced voice activity detection, wake word activation and instant transcription.| | 685|qunash/chatgpt-advanced !2025-03-2864910 | A browser extension that augments your ChatGPT prompts with web results.| | 686|Licoy/ChatGPT-Midjourney !2025-03-2864850|🎨 Own your own ChatGPT+Midjourney web service with one click| | 687|friuns2/BlackFriday-GPTs-Prompts !2025-03-2864744|List of free GPTs that doesn't require plus subscription| | 688|PixarAnimationStudios/OpenUSD !2025-03-2864700|Universal Scene Description| | 689|linyiLYi/street-fighter-ai !2025-03-2864630 |This is an AI agent for Street Fighter II Champion Edition.| | 690|run-llama/rags !2025-03-2864380|Build ChatGPT over your data, all with natural language| | 691|frdel/agent-zero !2025-03-2864154|Agent Zero AI framework| | 692|microsoft/DeepSpeedExamples !2025-03-2863911 |Example models using DeepSpeed| | 693|k8sgpt-ai/k8sgpt !2025-03-2863882|Giving Kubernetes Superpowers to everyone| | 694|open-metadata/OpenMetadata !2025-03-2863514|OpenMetadata is a unified platform for discovery, observability, and governance powered by a central metadata repository, in-depth lineage, and seamless team collaboration.| | 695|google/gemma.cpp !2025-03-2863163|lightweight, standalone C++ inference engine for Google's Gemma models.| | 696|RayVentura/ShortGPT !2025-03-286314-1|🚀🎬 ShortGPT - An experimental AI framework for automated short/video content creation. Enables creators to rapidly produce, manage, and deliver content using AI and automation.| | 697|openai/consistencymodels !2025-03-2862940 |Official repo for consistency models.| | 698|yangjianxin1/Firefly !2025-03-2862924|Firefly: Chinese conversational large language model (full-scale fine-tuning + QLoRA), supporting fine-tuning of Llma2, Llama, Baichuan, InternLM, Ziya, Bloom, and other large models| | 699|enricoros/big-AGI !2025-03-2862665|Generative AI suite powered by state-of-the-art models and providing advanced AI/AGI functions. It features AI personas, AGI functions, multi-model chats, text-to-image, voice, response streaming, code highlighting and execution, PDF import, presets for developers, much more. Deploy on-prem or in the cloud.| | 700|aptos-labs/aptos-core !2025-03-2862633|Aptos is a layer 1 blockchain built to support the widespread use of blockchain through better technology and user experience.| | 701|wenda-LLM/wenda !2025-03-286262-1 |Wenda: An LLM invocation platform. Its objective is to achieve efficient content generation tailored to specific environments while considering the limited computing resources of individuals and small businesses, as well as knowledge security and privacy concerns| | 702|Project-MONAI/MONAI !2025-03-2862603|AI Toolkit for Healthcare Imaging| | 703|HVision-NKU/StoryDiffusion !2025-03-2862470|Create Magic Story!| | 704|deepseek-ai/DeepSeek-LLM !2025-03-2862463|DeepSeek LLM: Let there be answers| | 705|Tohrusky/Final2x !2025-03-2862393|2^x Image Super-Resolution| | 706|OpenSPG/KAG !2025-03-28619611|KAG is a logical form-guided reasoning and retrieval framework based on OpenSPG engine and LLMs. It is used to build logical reasoning and factual Q&A solutions for professional domain knowledge bases. It can effectively overcome the shortcomings of the traditional RAG vector similarity calculation model.| | 707|Moonvy/OpenPromptStudio !2025-03-2861861 |AIGC Hint Word Visualization Editor| | 708|levihsu/OOTDiffusion !2025-03-2861761|Official implementation of OOTDiffusion| | 709|tmc/langchaingo !2025-03-2861729|LangChain for Go, the easiest way to write LLM-based programs in Go| | 710|vladmandic/automatic !2025-03-2861374|SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models| | 711|clovaai/donut !2025-03-2861231 |Official Implementation of OCR-free Document Understanding Transformer (Donut) and Synthetic Document Generator (SynthDoG), ECCV 2022| | 712|Shaunwei/RealChar !2025-03-286121-1|🎙️🤖Create, Customize and Talk to your AI Character/Companion in Realtime(All in One Codebase!). Have a natural seamless conversation with AI everywhere(mobile, web and terminal) using LLM OpenAI GPT3.5/4, Anthropic Claude2, Chroma Vector DB, Whisper Speech2Text, ElevenLabs Text2Speech🎙️🤖| | 713|microsoft/TinyTroupe !2025-03-2861142|LLM-powered multiagent persona simulation for imagination enhancement and business insights.| | 714| rustformers/llm !2025-03-2861010 | Run inference for Large Language Models on CPU, with Rust| | 715|firebase/firebase-ios-sdk !2025-03-2860950|Firebase SDK for Apple App Development| | 716|vespa-engine/vespa !2025-03-2860824|The open big data serving engine. https://vespa.ai| | 717|n4ze3m/page-assist !2025-03-28607610|Use your locally running AI models to assist you in your web browsing| | 718|Dooy/chatgpt-web-midjourney-proxy !2025-03-2860646|chatgpt web, midjourney, gpts,tts, whisper 一套ui全搞定| | 719|ethereum-optimism/optimism !2025-03-2860213|Optimism is Ethereum, scaled.| | 720|sczhou/ProPainter !2025-03-2859971|[ICCV 2023] ProPainter: Improving Propagation and Transformer for Video Inpainting| | 721|MineDojo/Voyager !2025-03-2859951 |An Open-Ended Embodied Agent with Large Language Models| | 722|lavague-ai/LaVague !2025-03-2859800|Automate automation with Large Action Model framework| | 723|SevaSk/ecoute !2025-03-2859770 |Ecoute is a live transcription tool that provides real-time transcripts for both the user's microphone input (You) and the user's speakers output (Speaker) in a textbox. It also generates a suggested response using OpenAI's GPT-3.5 for the user to say based on the live transcription of the conversation.| | 724|google/mesop !2025-03-2859661|| | 725|pengxiao-song/LaWGPT !2025-03-2859542 |Repo for LaWGPT, Chinese-Llama tuned with Chinese Legal knowledge| | 726|fr0gger/Awesome-GPT-Agents !2025-03-2859434|A curated list of GPT agents for cybersecurity| | 727|google-deepmind/graphcast !2025-03-2859412|| | 728|comet-ml/opik !2025-03-28594126|Open-source end-to-end LLM Development Platform| | 729|SciPhi-AI/R2R !2025-03-28594033|A framework for rapid development and deployment of production-ready RAG systems| | 730|SkalskiP/courses !2025-03-2859272 |This repository is a curated collection of links to various courses and resources about Artificial Intelligence (AI)| | 731|QuivrHQ/MegaParse !2025-03-2859122|File Parser optimised for LLM Ingestion with no loss 🧠 Parse PDFs, Docx, PPTx in a format that is ideal for LLMs.| | 732|pytorch-labs/gpt-fast !2025-03-2858971|Simple and efficient pytorch-native transformer text generation in !2025-03-2858886|Curated list of chatgpt prompts from the top-rated GPTs in the GPTs Store. Prompt Engineering, prompt attack & prompt protect. Advanced Prompt Engineering papers.| | 734|nilsherzig/LLocalSearch !2025-03-2858852|LLocalSearch is a completely locally running search aggregator using LLM Agents. The user can ask a question and the system will use a chain of LLMs to find the answer. The user can see the progress of the agents and the final answer. No OpenAI or Google API keys are needed.| | 735|kuafuai/DevOpsGPT !2025-03-285874-2|Multi agent system for AI-driven software development. Convert natural language requirements into working software. Supports any development language and extends the existing base code.| | 736|myshell-ai/MeloTTS !2025-03-2858486|High-quality multi-lingual text-to-speech library by MyShell.ai. Support English, Spanish, French, Chinese, Japanese and Korean.| | 737|OpenGVLab/LLaMA-Adapter !2025-03-2858421 |Fine-tuning LLaMA to follow Instructions within 1 Hour and 1.2M Parameters| | 738|volcengine/verl !2025-03-28582563|veRL: Volcano Engine Reinforcement Learning for LLM| | 739|a16z-infra/companion-app !2025-03-2858171|AI companions with memory: a lightweight stack to create and host your own AI companions| | 740|HumanAIGC/OutfitAnyone !2025-03-285816-1|Outfit Anyone: Ultra-high quality virtual try-on for Any Clothing and Any Person| | 741|josStorer/RWKV-Runner !2025-03-2857472|A RWKV management and startup tool, full automation, only 8MB. And provides an interface compatible with the OpenAI API. RWKV is a large language model that is fully open source and available for commercial use.| | 742|648540858/wvp-GB28181-pro !2025-03-2857414|WEB VIDEO PLATFORM是一个基于GB28181-2016标准实现的网络视频平台，支持NAT穿透，支持海康、大华、宇视等品牌的IPC、NVR、DVR接入。支持国标级联，支持rtsp/rtmp等视频流转发到国标平台，支持rtsp/rtmp等推流转发到国标平台。| | 743|ToonCrafter/ToonCrafter !2025-03-2857345|a research paper for generative cartoon interpolation| | 744|PawanOsman/ChatGPT !2025-03-2857191|OpenAI API Free Reverse Proxy| | 745|apache/hudi !2025-03-2857091|Upserts, Deletes And Incremental Processing on Big Data.| | 746| nsarrazin/serge !2025-03-2857081 | A web interface for chatting with Alpaca through llama.cpp. Fully dockerized, with an easy to use API| | 747|homanp/superagent !2025-03-2857021|🥷 Superagent - Build, deploy, and manage LLM-powered agents| | 748|ramonvc/freegpt-webui !2025-03-2856910|GPT 3.5/4 with a Chat Web UI. No API key is required.| | 749|baichuan-inc/baichuan-7B !2025-03-2856901|A large-scale 7B pretraining language model developed by BaiChuan-Inc.| | 750|Azure/azure-sdk-for-net !2025-03-2856792|This repository is for active development of the Azure SDK for .NET. For consumers of the SDK we recommend visiting our public developer docs at https://learn.microsoft.com/dotnet/azure/ or our versioned developer docs at https://azure.github.io/azure-sdk-for-net.| | 751|mnotgod96/AppAgent !2025-03-2856643|AppAgent: Multimodal Agents as Smartphone Users, an LLM-based multimodal agent framework designed to operate smartphone apps.| | 752|microsoft/TaskWeaver !2025-03-2856243|A code-first agent framework for seamlessly planning and executing data analytics tasks.| | 753| yetone/bob-plugin-openai-translator !2025-03-285600-1 | A Bob Plugin base ChatGPT API | | 754|PrefectHQ/marvin !2025-03-2855840 |A batteries-included library for building AI-powered software| | 755|microsoft/promptbase !2025-03-2855832|All things prompt engineering| | 756|fullstackhero/dotnet-starter-kit !2025-03-2855560|Production Grade Cloud-Ready .NET 8 Starter Kit (Web API + Blazor Client) with Multitenancy Support, and Clean/Modular Architecture that saves roughly 200+ Development Hours! All Batteries Included.| | 757|deepseek-ai/DeepSeek-Coder-V2 !2025-03-2855435|DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence| | 758|aiwaves-cn/agents !2025-03-2855391|An Open-source Framework for Autonomous Language Agents| | 759|microsoft/Mastering-GitHub-Copilot-for-Paired-Programming !2025-03-2855158|A 6 Lesson course teaching everything you need to know about harnessing GitHub Copilot and an AI Paired Programing resource.| | 760|allenai/OLMo !2025-03-2854506|Modeling, training, eval, and inference code for OLMo| | 761|apify/crawlee-python !2025-03-2854493|Crawlee—A web scraping and browser automation library for Python to build reliable crawlers. Extract data for AI, LLMs, RAG, or GPTs. Download HTML, PDF, JPG, PNG, and other files from websites. Works with BeautifulSoup, Playwright, and raw HTTP. Both headful and headless mode. With proxy rotation.| | 762|k2-fsa/sherpa-onnx !2025-03-28541520|Speech-to-text, text-to-speech, and speaker recongition using next-gen Kaldi with onnxruntime without Internet connection. Support embedded systems, Android, iOS, Raspberry Pi, RISC-V, x86_64 servers, websocket server/client, C/C++, Python, Kotlin, C#, Go, NodeJS, Java, Swift| | 763|TEN-framework/TEN-Agent !2025-03-28541411|TEN Agent is a realtime conversational AI agent powered by TEN. It seamlessly integrates the OpenAI Realtime API, RTC capabilities, and advanced features like weather updates, web search, computer vision, and Retrieval-Augmented Generation (RAG).| | 764|google/gemmapytorch !2025-03-2854010|The official PyTorch implementation of Google's Gemma models| | 765|snakers4/silero-vad !2025-03-2853858|Silero VAD: pre-trained enterprise-grade Voice Activity Detector| | 766|livekit/agents !2025-03-2853836|Build real-time multimodal AI applications 🤖🎙️📹| | 767|pipecat-ai/pipecat !2025-03-28537811|Open Source framework for voice and multimodal conversational AI| | 768|EricLBuehler/mistral.rs !2025-03-28536324|Blazingly fast LLM inference.| | 769|asg017/sqlite-vec !2025-03-28535810|Work-in-progress vector search SQLite extension that runs anywhere.| | 770|albertan017/LLM4Decompile !2025-03-2853563|Reverse Engineering: Decompiling Binary Code with Large Language Models| | 771|Permify/permify !2025-03-2853235|An open-source authorization as a service inspired by Google Zanzibar, designed to build and manage fine-grained and scalable authorization systems for any application.| | 772|imoneoi/openchat !2025-03-2853171|OpenChat: Advancing Open-source Language Models with Imperfect Data| | 773|mosaicml/composer !2025-03-2853140|Train neural networks up to 7x faster| | 774|dsdanielpark/Bard-API !2025-03-285277-1 |The python package that returns a response of Google Bard through API.| | 775|lxfater/inpaint-web !2025-03-2852552|A free and open-source inpainting & image-upscaling tool powered by webgpu and wasm on the browser。| | 776|leanprover/lean4 !2025-03-2852441|Lean 4 programming language and theorem prover| | 777|AILab-CVC/YOLO-World !2025-03-2852415|Real-Time Open-Vocabulary Object Detection| | 778|openchatai/OpenChat !2025-03-2852260 |Run and create custom ChatGPT-like bots with OpenChat, embed and share these bots anywhere, the open-source chatbot console.| | 779|mufeedvh/code2prompt !2025-03-28519414|A CLI tool to convert your codebase into a single LLM prompt with source tree, prompt templating, and token counting.| | 780|biobootloader/wolverine !2025-03-2851700 |Automatically repair python scripts through GPT-4 to give them regenerative abilities.| | 781|huggingface/parler-tts !2025-03-2851671|Inference and training library for high-quality TTS models.| | 782|Akegarasu/lora-scripts !2025-03-2851308 |LoRA training scripts use kohya-ss's trainer, for diffusion model.| | 783|openchatai/OpenCopilot !2025-03-285128-3|🤖 🔥 Let your users chat with your product features and execute things by text - open source Shopify sidekick| | 784|e2b-dev/fragments !2025-03-2851228|Open-source Next.js template for building apps that are fully generated by AI. By E2B.| | 785|microsoft/SynapseML !2025-03-2851132|Simple and Distributed Machine Learning| | 786|aigc-apps/sd-webui-EasyPhoto !2025-03-285108-1|📷 EasyPhoto | | 787|ChaoningZhang/MobileSAM !2025-03-2850944|This is the official code for Faster Segment Anything (MobileSAM) project that makes SAM lightweight| | 788|huggingface/alignment-handbook !2025-03-2850932|Robust recipes for to align language models with human and AI preferences| | 789|alpkeskin/mosint !2025-03-2850920|An automated e-mail OSINT tool| | 790|TaskingAI/TaskingAI !2025-03-2850891|The open source platform for AI-native application development.| | 791|lipku/metahuman-stream !2025-03-28507615|Real time interactive streaming digital human| | 792|OpenInterpreter/01 !2025-03-2850530|The open-source language model computer| | 793|open-compass/opencompass !2025-03-28505111|OpenCompass is an LLM evaluation platform, supporting a wide range of models (InternLM2,GPT-4,LLaMa2, Qwen,GLM, Claude, etc) over 100+ datasets.| | 794|xxlong0/Wonder3D !2025-03-2850491|A cross-domain diffusion model for 3D reconstruction from a single image| | 795|pytorch/torchtune !2025-03-2850342|A Native-PyTorch Library for LLM Fine-tuning| | 796|SuperDuperDB/superduperdb !2025-03-2850192|🔮 SuperDuperDB: Bring AI to your database: Integrate, train and manage any AI models and APIs directly with your database and your data.| | 797|WhiskeySockets/Baileys !2025-03-2850057|Lightweight full-featured typescript/javascript WhatsApp Web API| | 798| mpociot/chatgpt-vscode !2025-03-2849890 | A VSCode extension that allows you to use ChatGPT | | 799|OpenGVLab/DragGAN !2025-03-2849880|Unofficial Implementation of DragGAN - "Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold" （DragGAN 全功能实现，在线Demo，本地部署试用，代码、模型已全部开源，支持Windows, macOS, Linux）| | 800|microsoft/LLMLingua !2025-03-2849824|To speed up LLMs' inference and enhance LLM's perceive of key information, compress the prompt and KV-Cache, which achieves up to 20x compression with minimal performance loss.| | 801|Zipstack/unstract !2025-03-2849745|No-code LLM Platform to launch APIs and ETL Pipelines to structure unstructured documents| | 802|OpenBMB/ToolBench !2025-03-2849621|An open platform for training, serving, and evaluating large language model for tool learning.| | 803|Fanghua-Yu/SUPIR !2025-03-2849593|SUPIR aims at developing Practical Algorithms for Photo-Realistic Image Restoration In the Wild| | 804|GaiaNet-AI/gaianet-node !2025-03-2849360|Install and run your own AI agent service| | 805|qodo-ai/qodo-cover !2025-03-284922-1|Qodo-Cover: An AI-Powered Tool for Automated Test Generation and Code Coverage Enhancement! 💻🤖🧪🐞| | 806|Zejun-Yang/AniPortrait !2025-03-2849042|AniPortrait: Audio-Driven Synthesis of Photorealistic Portrait Animation| | 807|lvwzhen/law-cn-ai !2025-03-2848901 |⚖️ AI Legal Assistant| | 808|developersdigest/llm-answer-engine !2025-03-2848740|Build a Perplexity-Inspired Answer Engine Using Next.js, Groq, Mixtral, Langchain, OpenAI, Brave & Serper| | 809|Plachtaa/VITS-fast-fine-tuning !2025-03-2848640|This repo is a pipeline of VITS finetuning for fast speaker adaptation TTS, and many-to-many voice conversion| | 810|espeak-ng/espeak-ng !2025-03-2848601|eSpeak NG is an open source speech synthesizer that supports more than hundred languages and accents.| | 811|ant-research/CoDeF !2025-03-2848581|[CVPR'24 Highlight] Official PyTorch implementation of CoDeF: Content Deformation Fields for Temporally Consistent Video Processing| | 812|deepseek-ai/DeepSeek-V2 !2025-03-2848512|| | 813|XRPLF/rippled !2025-03-2848210|Decentralized cryptocurrency blockchain daemon implementing the XRP Ledger protocol in C++| | 814|AutoMQ/automq !2025-03-28478721|AutoMQ is a cloud-first alternative to Kafka by decoupling durability to S3 and EBS. 10x cost-effective. Autoscale in seconds. Single-digit ms latency.| | 815|AILab-CVC/VideoCrafter !2025-03-2847800|VideoCrafter1: Open Diffusion Models for High-Quality Video Generation| | 816|nautechsystems/nautilustrader !2025-03-2847702|A high-performance algorithmic trading platform and event-driven backtester| | 817|kyegomez/swarms !2025-03-2847563|The Enterprise-Grade Production-Ready Multi-Agent Orchestration Framework Join our Community: https://discord.com/servers/agora-999382051935506503| | 818|Deci-AI/super-gradients !2025-03-2847310 |Easily train or fine-tune SOTA computer vision models with one open source training library. The home of Yolo-NAS.| | 819|QwenLM/Qwen2.5-Coder !2025-03-2847236|Qwen2.5-Coder is the code version of Qwen2.5, the large language model series developed by Qwen team, Alibaba Cloud.| | 820|SCIR-HI/Huatuo-Llama-Med-Chinese !2025-03-2847191 |Repo for HuaTuo (华驼), Llama-7B tuned with Chinese medical knowledge| | 821|togethercomputer/RedPajama-Data !2025-03-2846841 |code for preparing large datasets for training large language models| | 822|mishushakov/llm-scraper !2025-03-2846704|Turn any webpage into structured data using LLMs| | 823|1rgs/jsonformer !2025-03-2846663 |A Bulletproof Way to Generate Structured JSON from Language Models| | 824|anti-work/shortest !2025-03-2846565|QA via natural language AI tests| | 825|dnhkng/GlaDOS !2025-03-2846510|This is the Personality Core for GLaDOS, the first steps towards a real-life implementation of the AI from the Portal series by Valve.| | 826|Nukem9/dlssg-to-fsr3 !2025-03-2846380|Adds AMD FSR3 Frame Generation to games by replacing Nvidia DLSS-G Frame Generation (nvngx_dlssg).| | 827|BuilderIO/ai-shell !2025-03-2846373 |A CLI that converts natural language to shell commands.| | 828|facebookincubator/AITemplate !2025-03-2846220 |AITemplate is a Python framework which renders neural network into high performance CUDA/HIP C++ code. Specialized for FP16 TensorCore (NVIDIA GPU) and MatrixCore (AMD GPU) inference.| | 829|terraform-aws-modules/terraform-aws-eks !2025-03-2846030|Terraform module to create AWS Elastic Kubernetes (EKS) resources 🇺🇦| | 830|timescale/pgai !2025-03-2845915|A suite of tools to develop RAG, semantic search, and other AI applications more easily with PostgreSQL| | 831|awslabs/multi-agent-orchestrator !2025-03-2845788|Flexible and powerful framework for managing multiple AI agents and handling complex conversations| | 832|sanchit-gandhi/whisper-jax !2025-03-2845771 |Optimised JAX code for OpenAI's Whisper Model, largely built on the Hugging Face Transformers Whisper implementation| | 833|NVIDIA/NeMo-Guardrails !2025-03-2845755|NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.| | 834|PathOfBuildingCommunity/PathOfBuilding !2025-03-2845480|Offline build planner for Path of Exile.| | 835|UX-Decoder/Segment-Everything-Everywhere-All-At-Once !2025-03-2845412 |Official implementation of the paper "Segment Everything Everywhere All at Once"| | 836|build-trust/ockam !2025-03-2845171|Orchestrate end-to-end encryption, cryptographic identities, mutual authentication, and authorization policies between distributed applications – at massive scale.| | 837|google-research/timesfm !2025-03-2845135|TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.| | 838|luosiallen/latent-consistency-model !2025-03-2844842|Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference| | 839|NVlabs/neuralangelo !2025-03-2844740|Official implementation of "Neuralangelo: High-Fidelity Neural Surface Reconstruction" (CVPR 2023)| | 840|kyegomez/tree-of-thoughts !2025-03-2844720 |Plug in and Play Implementation of Tree of Thoughts: Deliberate Problem Solving with Large Language Models that Elevates Model Reasoning by atleast 70%| | 841|sjvasquez/handwriting-synthesis !2025-03-2844720 |Handwriting Synthesis with RNNs ✏️| | 842| madawei2699/myGPTReader !2025-03-2844420 | A slack bot that can read any webpage, ebook or document and summarize it with chatGPT | | 843|OpenBMB/AgentVerse !2025-03-2844413|🤖 AgentVerse 🪐 provides a flexible framework that simplifies the process of building custom multi-agent environments for large language models (LLMs).| | 844|argmaxinc/WhisperKit !2025-03-2844395|Swift native speech recognition on-device for iOS and macOS applications.| | 845|landing-ai/vision-agent !2025-03-2844346|Vision agent| | 846|InternLM/xtuner !2025-03-2844273|An efficient, flexible and full-featured toolkit for fine-tuning large models (InternLM, Llama, Baichuan, Qwen, ChatGLM)| | 847|google-deepmind/alphageometry !2025-03-284421-1|Solving Olympiad Geometry without Human Demonstrations| | 848|ostris/ai-toolkit !2025-03-2844093|Various AI scripts. Mostly Stable Diffusion stuff.| | 849|LLM-Red-Team/kimi-free-api !2025-03-2844004|🚀 KIMI AI 长文本大模型白嫖服务，支持高速流式输出、联网搜索、长文档解读、图像解析、多轮对话，零配置部署，多路token支持，自动清理会话痕迹。| | 850|argilla-io/argilla !2025-03-2843991|Argilla is a collaboration platform for AI engineers and domain experts that require high-quality outputs, full data ownership, and overall efficiency.| | 851|spring-projects/spring-ai !2025-03-28438419|An Application Framework for AI Engineering| | 852|alibaba-damo-academy/FunClip !2025-03-2843555|Open-source, accurate and easy-to-use video clipping tool, LLM based AI clipping intergrated | | 853|yisol/IDM-VTON !2025-03-2843541|IDM-VTON : Improving Diffusion Models for Authentic Virtual Try-on in the Wild| | 854|fchollet/ARC-AGI !2025-03-2843368|The Abstraction and Reasoning Corpus| | 855|MahmoudAshraf97/whisper-diarization !2025-03-2843064|Automatic Speech Recognition with Speaker Diarization based on OpenAI Whisper| | 856|Speykious/cve-rs !2025-03-2843047|Blazingly 🔥 fast 🚀 memory vulnerabilities, written in 100% safe Rust. 🦀| | 857|Blealtan/efficient-kan !2025-03-2842770|An efficient pure-PyTorch implementation of Kolmogorov-Arnold Network (KAN).| | 858|smol-ai/GodMode !2025-03-284249-1|AI Chat Browser: Fast, Full webapp access to ChatGPT / Claude / Bard / Bing / Llama2! I use this 20 times a day.| | 859|openai/plugins-quickstart !2025-03-284235-4 |Get a ChatGPT plugin up and running in under 5 minutes!| | 860|Doriandarko/maestro !2025-03-2842260|A framework for Claude Opus to intelligently orchestrate subagents.| | 861|philz1337x/clarity-upscaler !2025-03-2842204|Clarity-Upscaler: Reimagined image upscaling for everyone| | 862|facebookresearch/co-tracker !2025-03-2842142|CoTracker is a model for tracking any point (pixel) on a video.| | 863|xlang-ai/OpenAgents !2025-03-2842031|OpenAgents: An Open Platform for Language Agents in the Wild| | 864|alibaba/higress !2025-03-28419514|🤖 AI Gateway | | 865|ray-project/llm-numbers !2025-03-2841920 |Numbers every LLM developer should know| | 866|fudan-generative-vision/champ !2025-03-2841820|Champ: Controllable and Consistent Human Image Animation with 3D Parametric Guidance| | 867|NVIDIA/garak !2025-03-2841795|the LLM vulnerability scanner| | 868|leetcode-mafia/cheetah !2025-03-2841740 |Whisper & GPT-based app for passing remote SWE interviews| | 869|ragapp/ragapp !2025-03-2841710|The easiest way to use Agentic RAG in any enterprise| | 870|collabora/WhisperSpeech !2025-03-2841692|An Open Source text-to-speech system built by inverting Whisper.| | 871|Facico/Chinese-Vicuna !2025-03-2841520 |Chinese-Vicuna: A Chinese Instruction-following LLaMA-based Model| | 872|openai/grok !2025-03-2841381|| | 873|CrazyBoyM/llama3-Chinese-chat !2025-03-2841361|Llama3 Chinese Repository with modified versions, and training and deployment resources| | 874|luban-agi/Awesome-AIGC-Tutorials !2025-03-2841301|Curated tutorials and resources for Large Language Models, AI Painting, and more.| | 875|damo-vilab/AnyDoor !2025-03-2841192|Official implementations for paper: Anydoor: zero-shot object-level image customization| | 876|raspberrypi/pico-sdk !2025-03-2841072|| | 877|mshumer/gpt-llm-trainer !2025-03-284097-1|| | 878|metavoiceio/metavoice-src !2025-03-284076-1|AI for human-level speech intelligence| | 879|intelowlproject/IntelOwl !2025-03-2840763|IntelOwl: manage your Threat Intelligence at scale| | 880|a16z-infra/ai-getting-started !2025-03-2840682|A Javascript AI getting started stack for weekend projects, including image/text models, vector stores, auth, and deployment configs| | 881|MarkFzp/mobile-aloha !2025-03-2840641|Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation| | 882| keijiro/AICommand !2025-03-2840380 | ChatGPT integration with Unity Editor | | 883|Tencent/HunyuanDiT !2025-03-2840214|Hunyuan-DiT : A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding| | 884|hengyoush/kyanos !2025-03-2840061|Visualize the time packets spend in the kernel, watch & analyze in command line.| | 885|agiresearch/AIOS !2025-03-2840045|AIOS: LLM Agent Operating System| | 886|truefoundry/cognita !2025-03-2839773|RAG (Retrieval Augmented Generation) Framework for building modular, open source applications for production by TrueFoundry| | 887|X-PLUG/MobileAgent !2025-03-2839557|Mobile-Agent: Autonomous Multi-Modal Mobile Device Agent with Visual Perception| | 888|jackMort/ChatGPT.nvim !2025-03-2839231|ChatGPT Neovim Plugin: Effortless Natural Language Generation with OpenAI's ChatGPT API| | 889|microsoft/RD-Agent !2025-03-28388422|Research and development (R&D) is crucial for the enhancement of industrial productivity, especially in the AI era, where the core aspects of R&D are mainly focused on data and models. We are committed to automate these high-value generic R&D processes through our open source R&D automation tool RD-Agent, which let AI drive data-driven AI.| | 890|Significant-Gravitas/Auto-GPT-Plugins !2025-03-283882-1 |Plugins for Auto-GPT| | 891|apple/ml-mgie !2025-03-2838770|| | 892|OpenDriveLab/UniAD !2025-03-2838727|[CVPR 2023 Best Paper] Planning-oriented Autonomous Driving| | 893|llSourcell/DoctorGPT !2025-03-2838640|DoctorGPT is an LLM that can pass the US Medical Licensing Exam. It works offline, it's cross-platform, & your health data stays private.| | 894|FlagAI-Open/FlagAI !2025-03-2838601|FlagAI (Fast LArge-scale General AI models) is a fast, easy-to-use and extensible toolkit for large-scale model.| | 895|krishnaik06/Roadmap-To-Learn-Generative-AI-In-2024 !2025-03-2838513|Roadmap To Learn Generative AI In 2024| | 896|SysCV/sam-hq !2025-03-2838491|Segment Anything in High Quality| | 897|google/security-research !2025-03-2838420|This project hosts security advisories and their accompanying proof-of-concepts related to research conducted at Google which impact non-Google owned code.| | 898|shroominic/codeinterpreter-api !2025-03-2838330|Open source implementation of the ChatGPT Code Interpreter 👾| | 899|Yonom/assistant-ui !2025-03-2838308|React Components for AI Chat 💬 🚀| | 900|nucleuscloud/neosync !2025-03-2838262|Open source data anonymization and synthetic data orchestration for developers. Create high fidelity synthetic data and sync it across your environments.| | 901|ravenscroftj/turbopilot !2025-03-2838230 |Turbopilot is an open source large-language-model based code completion engine that runs locally on CPU| | 902|NVlabs/Sana !2025-03-28380810|SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer| | 903|huggingface/distil-whisper !2025-03-2838061|Distilled variant of Whisper for speech recognition. 6x faster, 50% smaller, within 1% word error rate.| | 904|Codium-ai/AlphaCodium !2025-03-2837971|code generation tool that surpasses most human competitors in CodeContests| | 905|fixie-ai/ultravox !2025-03-2837710|A fast multimodal LLM for real-time voice| | 906|unit-mesh/auto-dev !2025-03-28375715|🧙‍AutoDev: The AI-powered coding wizard with multilingual support 🌐, auto code generation 🏗️, and a helpful bug-slaying assistant 🐞! Customizable prompts 🎨 and a magic Auto Dev/Testing/Document/Agent feature 🧪 included! 🚀| | 907|Marker-Inc-Korea/AutoRAG !2025-03-2837432|AutoML tool for RAG| | 908|deepseek-ai/DeepSeek-VL !2025-03-283734-1|DeepSeek-VL: Towards Real-World Vision-Language Understanding| | 909|hiyouga/ChatGLM-Efficient-Tuning !2025-03-283692-1|Fine-tuning ChatGLM-6B with PEFT | | 910| Yue-Yang/ChatGPT-Siri !2025-03-2836921 | Shortcuts for Siri using ChatGPT API gpt-3.5-turbo model | | 911|0hq/WebGPT !2025-03-2836901 |Run GPT model on the browser with WebGPU. An implementation of GPT inference in less than ~2000 lines of vanilla Javascript.| | 912|cvg/LightGlue !2025-03-2836903|LightGlue: Local Feature Matching at Light Speed (ICCV 2023)| | 913|deanxv/coze-discord-proxy !2025-03-2836791|代理Discord-Bot对话Coze-Bot，实现API形式请求GPT4对话模型/微调模型| | 914|MervinPraison/PraisonAI !2025-03-2836764|PraisonAI application combines AutoGen and CrewAI or similar frameworks into a low-code solution for building and managing multi-agent LLM systems, focusing on simplicity, customisation, and efficient human-agent collaboration.| | 915|Ironclad/rivet !2025-03-2836345 |The open-source visual AI programming environment and TypeScript library| | 916|BasedHardware/OpenGlass !2025-03-2835851|Turn any glasses into AI-powered smart glasses| | 917|ricklamers/gpt-code-ui !2025-03-2835840 |An open source implementation of OpenAI's ChatGPT Code interpreter| | 918|whoiskatrin/chart-gpt !2025-03-2835830 |AI tool to build charts based on text input| | 919|github/CopilotForXcode !2025-03-2835788|Xcode extension for GitHub Copilot| | 920|hemansnation/God-Level-Data-Science-ML-Full-Stack !2025-03-2835570 |A collection of scientific methods, processes, algorithms, and systems to build stories & models. This roadmap contains 16 Chapters, whether you are a fresher in the field or an experienced professional who wants to transition into Data Science & AI| | 921|pytorch/torchchat !2025-03-2835461|Run PyTorch LLMs locally on servers, desktop and mobile| | 922| Kent0n-Li/ChatDoctor !2025-03-2835451 | A Medical Chat Model Fine-tuned on LLaMA Model using Medical Domain Knowledge | | 923|xtekky/chatgpt-clone !2025-03-283519-1 |ChatGPT interface with better UI| | 924|jupyterlab/jupyter-ai !2025-03-2835120|A generative AI extension for JupyterLab| | 925|pytorch/torchtitan !2025-03-2835064|A native PyTorch Library for large model training| | 926|minimaxir/simpleaichat !2025-03-2835031|Python package for easily interfacing with chat apps, with robust features and minimal code complexity.| | 927|srush/Tensor-Puzzles !2025-03-2834930|Solve puzzles. Improve your pytorch.| | 928|Helicone/helicone !2025-03-2834918|🧊 Open source LLM-Observability Platform for Developers. One-line integration for monitoring, metrics, evals, agent tracing, prompt management, playground, etc. Supports OpenAI SDK, Vercel AI SDK, Anthropic SDK, LiteLLM, LLamaIndex, LangChain, and more. 🍓 YC W23| | 929|run-llama/llama-hub !2025-03-2834740|A library of data loaders for LLMs made by the community -- to be used with LlamaIndex and/or LangChain| | 930|NExT-GPT/NExT-GPT !2025-03-2834700|Code and models for NExT-GPT: Any-to-Any Multimodal Large Language Model| | 931|souzatharsis/podcastfy !2025-03-2834661|An Open Source Python alternative to NotebookLM's podcast feature: Transforming Multimodal Content into Captivating Multilingual Audio Conversations with GenAI| | 932|Dataherald/dataherald !2025-03-2834450|Interact with your SQL database, Natural Language to SQL using LLMs| | 933|iryna-kondr/scikit-llm !2025-03-2834350 |Seamlessly integrate powerful language models like ChatGPT into scikit-learn for enhanced text analysis tasks.| | 934|Netflix/maestro !2025-03-2834230|Maestro: Netflix’s Workflow Orchestrator| | 935|CanadaHonk/porffor !2025-03-2833560|A from-scratch experimental AOT JS engine, written in JS| | 936|hustvl/Vim !2025-03-2833323|Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model| | 937|pashpashpash/vault-ai !2025-03-2833250 |OP Vault ChatGPT: Give ChatGPT long-term memory using the OP Stack (OpenAI + Pinecone Vector Database). Upload your own custom knowledge base files (PDF, txt, etc) using a simple React frontend.| | 938|tencentmusic/supersonic !2025-03-28330611|SuperSonic is the next-generation BI platform that integrates Chat BI (powered by LLM) and Headless BI (powered by semantic layer) paradigms.| | 939|billmei/every-chatgpt-gui !2025-03-2832981|Every front-end GUI client for ChatGPT| | 940|microsoft/torchgeo !2025-03-2832772|TorchGeo: datasets, samplers, transforms, and pre-trained models for geospatial data| | 941|LLMBook-zh/LLMBook-zh.github.io !2025-03-28326110|《大语言模型》作者：赵鑫，李军毅，周昆，唐天一，文继荣| | 942|dvlab-research/MiniGemini !2025-03-2832601|Official implementation for Mini-Gemini| | 943|rashadphz/farfalle !2025-03-2832460|🔍 AI search engine - self-host with local or cloud LLMs| | 944|Luodian/Otter !2025-03-2832450|🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.| | 945|AprilNEA/ChatGPT-Admin-Web !2025-03-2832370 | ChatGPT WebUI with user management and admin dashboard system| | 946|MarkFzp/act-plus-plus !2025-03-2832365|Imitation Learning algorithms with Co-traing for Mobile ALOHA: ACT, Diffusion Policy, VINN| | 947|ethen8181/machine-learning !2025-03-2832310|🌎 machine learning tutorials (mainly in Python3)| | 948|opengeos/segment-geospatial !2025-03-2832312 |A Python package for segmenting geospatial data with the Segment Anything Model (SAM)| | 949|iusztinpaul/hands-on-llms !2025-03-283225-2|🦖 𝗟𝗲𝗮𝗿𝗻 about 𝗟𝗟𝗠𝘀, 𝗟𝗟𝗠𝗢𝗽𝘀, and 𝘃𝗲𝗰𝘁𝗼𝗿 𝗗𝗕𝘀 for free by designing, training, and deploying a real-time financial advisor LLM system ~ 𝘴𝘰𝘶𝘳𝘤𝘦 𝘤𝘰𝘥𝘦 + 𝘷𝘪𝘥𝘦𝘰 & 𝘳𝘦𝘢𝘥𝘪𝘯𝘨 𝘮𝘢𝘵𝘦𝘳𝘪𝘢𝘭𝘴| | 950|ToTheBeginning/PuLID !2025-03-2832221|Official code for PuLID: Pure and Lightning ID Customization via Contrastive Alignment| | 951|neo4j-labs/llm-graph-builder !2025-03-2832164|Neo4j graph construction from unstructured data using LLMs| | 952|OpenGVLab/InternGPT !2025-03-2832150 |InternGPT (iGPT) is an open source demo platform where you can easily showcase your AI models. Now it supports DragGAN, ChatGPT, ImageBind, multimodal chat like GPT-4, SAM, interactive image editing, etc. Try it at igpt.opengvlab.com (支持DragGAN、ChatGPT、ImageBind、SAM的在线Demo系统)| | 953|PKU-YuanGroup/Video-LLaVA !2025-03-2832060 |Video-LLaVA: Learning United Visual Representation by Alignment Before Projection| | 954|DataTalksClub/llm-zoomcamp !2025-03-2832030|LLM Zoomcamp - a free online course about building an AI bot that can answer questions about your knowledge base| | 955|gptscript-ai/gptscript !2025-03-2832010|Natural Language Programming| |!green-up-arrow.svg 956|isaac-sim/IsaacLab !2025-03-28320113|Unified framework for robot learning built on NVIDIA Isaac Sim| |!red-down-arrow 957|ai-boost/Awesome-GPTs !2025-03-2832003|Curated list of awesome GPTs 👍.| | 958|huggingface/safetensors !2025-03-2831901|Simple, safe way to store and distribute tensors| | 959|linyiLYi/bilibot !2025-03-2831771|A local chatbot fine-tuned by bilibili user comments.| | 960| project-baize/baize-chatbot !2025-03-283168-1 | Let ChatGPT teach your own chatbot in hours with a single GPU! | | 961|Azure-Samples/cognitive-services-speech-sdk !2025-03-2831280|Sample code for the Microsoft Cognitive Services Speech SDK| | 962|microsoft/Phi-3CookBook !2025-03-2831231|This is a Phi-3 book for getting started with Phi-3. Phi-3, a family of open AI models developed by Microsoft. Phi-3 models are the most capable and cost-effective small language models (SLMs) available, outperforming models of the same size and next size up across a variety of language, reasoning, coding, and math benchmarks.| | 963|neuralmagic/deepsparse !2025-03-2831180|Sparsity-aware deep learning inference runtime for CPUs| | 964|sugarforever/chat-ollama !2025-03-2831000|ChatOllama is an open source chatbot based on LLMs. It supports a wide range of language models, and knowledge base management.| | 965|amazon-science/chronos-forecasting !2025-03-2830974|Chronos: Pretrained (Language) Models for Probabilistic Time Series Forecasting| | 966|damo-vilab/i2vgen-xl !2025-03-2830902|Official repo for VGen: a holistic video generation ecosystem for video generation building on diffusion models| | 967|google-deepmind/gemma !2025-03-2830733|Open weights LLM from Google DeepMind.| | 968|iree-org/iree !2025-03-2830733|A retargetable MLIR-based machine learning compiler and runtime toolkit.| | 969|NVlabs/VILA !2025-03-2830724|VILA - a multi-image visual language model with training, inference and evaluation recipe, deployable from cloud to edge (Jetson Orin and laptops)| | 970|microsoft/torchscale !2025-03-2830661|Foundation Architecture for (M)LLMs| | 971|openai/openai-realtime-console !2025-03-2830656|React app for inspecting, building and debugging with the Realtime API| | 972|daveshap/OpenAIAgentSwarm !2025-03-2830610|HAAS = Hierarchical Autonomous Agent Swarm - "Resistance is futile!"| | 973|microsoft/PromptWizard !2025-03-2830555|Task-Aware Agent-driven Prompt Optimization Framework| | 974|CVI-SZU/Linly !2025-03-2830490 |Chinese-LLaMA basic model; ChatFlow Chinese conversation model; NLP pre-training/command fine-tuning dataset| | 975|cohere-ai/cohere-toolkit !2025-03-2830130|Toolkit is a collection of prebuilt components enabling users to quickly build and deploy RAG applications.| | 976|adamcohenhillel/ADeus !2025-03-2830131|An open source AI wearable device that captures what you say and hear in the real world and then transcribes and stores it on your own server. You can then chat with Adeus using the app, and it will have all the right context about what you want to talk about - a truly personalized, personal AI.| | 977|Lightning-AI/LitServe !2025-03-2830132|Lightning-fast serving engine for AI models. Flexible. Easy. Enterprise-scale.| | 978|potpie-ai/potpie !2025-03-2829973|Prompt-To-Agent : Create custom engineering agents for your codebase| | 979|ant-design/x !2025-03-28299529|Craft AI-driven interfaces effortlessly 🤖| | 980|meta-llama/PurpleLlama !2025-03-2829832|Set of tools to assess and improve LLM security.| | 981|williamyang1991/RerenderAVideo !2025-03-2829800|[SIGGRAPH Asia 2023] Rerender A Video: Zero-Shot Text-Guided Video-to-Video Translation| | 982|baichuan-inc/Baichuan-13B !2025-03-2829790|A 13B large language model developed by Baichuan Intelligent Technology| | 983|Stability-AI/stable-audio-tools !2025-03-2829761|Generative models for conditional audio generation| | 984|li-plus/chatglm.cpp !2025-03-2829720|C++ implementation of ChatGLM-6B & ChatGLM2-6B & ChatGLM3 & more LLMs| | 985|NVIDIA/GenerativeAIExamples !2025-03-2829546|Generative AI reference workflows optimized for accelerated infrastructure and microservice architecture.| | 986|Josh-XT/AGiXT !2025-03-2829521 |AGiXT is a dynamic AI Automation Platform that seamlessly orchestrates instruction management and complex task execution across diverse AI providers. Combining adaptive memory, smart features, and a versatile plugin system, AGiXT delivers efficient and comprehensive AI solutions.| | 987|MrForExample/ComfyUI-3D-Pack !2025-03-2829515|An extensive node suite that enables ComfyUI to process 3D inputs (Mesh & UV Texture, etc) using cutting edge algorithms (3DGS, NeRF, etc.)| | 988|olimorris/codecompanion.nvim !2025-03-28295111|✨ AI-powered coding, seamlessly in Neovim. Supports Anthropic, Copilot, Gemini, Ollama, OpenAI and xAI LLMs| | 989|salesforce/CodeT5 !2025-03-282940-1 |Home of CodeT5: Open Code LLMs for Code Understanding and Generation| | 990|facebookresearch/ijepa !2025-03-2829391|Official codebase for I-JEPA, the Image-based Joint-Embedding Predictive Architecture. First outlined in the CVPR paper, "Self-supervised learning from images with a joint-embedding predictive architecture."| | 991|eureka-research/Eureka !2025-03-2829351|Official Repository for "Eureka: Human-Level Reward Design via Coding Large Language Models"| | 992|NVIDIA/trt-llm-rag-windows !2025-03-282934-1|A developer reference project for creating Retrieval Augmented Generation (RAG) chatbots on Windows using TensorRT-LLM| | 993|gmpetrov/databerry !2025-03-282930-1|The no-code platform for building custom LLM Agents| | 994|AI4Finance-Foundation/FinRobot !2025-03-28291946|FinRobot: An Open-Source AI Agent Platform for Financial Applications using LLMs 🚀 🚀 🚀| | 995|nus-apr/auto-code-rover !2025-03-2829013|A project structure aware autonomous software engineer aiming for autonomous program improvement| | 996|deepseek-ai/DreamCraft3D !2025-03-2828921|[ICLR 2024] Official implementation of DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior| | 997|mlabonne/llm-datasets !2025-03-2828848|High-quality datasets, tools, and concepts for LLM fine-tuning.| | 998|facebookresearch/jepa !2025-03-2828712|PyTorch code and models for V-JEPA self-supervised learning from video.| | 999|facebookresearch/habitat-sim !2025-03-2828604|A flexible, high-performance 3D simulator for Embodied AI research.| | 1000|xenova/whisper-web !2025-03-2828581|ML-powered speech recognition directly in your browser| | 1001|cvlab-columbia/zero123 !2025-03-2828530|Zero-1-to-3: Zero-shot One Image to 3D Object: https://zero123.cs.columbia.edu/| | 1002|yuruotong1/autoMate !2025-03-28285121|Like Manus, Computer Use Agent(CUA) and Omniparser, we are computer-using agents.AI-driven local automation assistant that uses natural language to make computers work by themselves| | 1003|muellerberndt/mini-agi !2025-03-282845-1 |A minimal generic autonomous agent based on GPT3.5/4. Can analyze stock prices, perform network security tests, create art, and order pizza.| | 1004|allenai/open-instruct !2025-03-2828432|| | 1005|CodingChallengesFYI/SharedSolutions !2025-03-2828360|Publicly shared solutions to Coding Challenges| | 1006|hegelai/prompttools !2025-03-2828220|Open-source tools for prompt testing and experimentation, with support for both LLMs (e.g. OpenAI, LLaMA) and vector databases (e.g. Chroma, Weaviate).| | 1007|mazzzystar/Queryable !2025-03-2828222|Run CLIP on iPhone to Search Photos.| | 1008|Doubiiu/DynamiCrafter !2025-03-2828173|DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors| | 1009|SamurAIGPT/privateGPT !2025-03-282805-1 |An app to interact privately with your documents using the power of GPT, 100% privately, no data leaks| | 1010|facebookresearch/Pearl !2025-03-2827951|A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta.| | 1011|intuitem/ciso-assistant-community !2025-03-2827954|CISO Assistant is a one-stop-shop for GRC, covering Risk, AppSec and Audit Management and supporting +70 frameworks worldwide with auto-mapping: NIST CSF, ISO 27001, SOC2, CIS, PCI DSS, NIS2, CMMC, PSPF, GDPR, HIPAA, Essential Eight, NYDFS-500, DORA, NIST AI RMF, 800-53, 800-171, CyFun, CJIS, AirCyber, NCSC, ECC, SCF and so much more| | 1012|facebookresearch/audio2photoreal !2025-03-2827840|Code and dataset for photorealistic Codec Avatars driven from audio| | 1013|Azure/azure-rest-api-specs !2025-03-2827770|The source for REST API specifications for Microsoft Azure.| | 1014|SCUTlihaoyu/open-chat-video-editor !2025-03-2827690 |Open source short video automatic generation tool| | 1015|Alpha-VLLM/LLaMA2-Accessory !2025-03-2827642|An Open-source Toolkit for LLM Development| | 1016|johnma2006/mamba-minimal !2025-03-2827601|Simple, minimal implementation of the Mamba SSM in one file of PyTorch.| | 1017|nerfstudio-project/gsplat !2025-03-2827576|CUDA accelerated rasterization of gaussian splatting| | 1018|Physical-Intelligence/openpi !2025-03-28274617|| | 1019|leptonai/leptonai !2025-03-2827246|A Pythonic framework to simplify AI service building| |!green-up-arrow.svg 1020|joanrod/star-vector !2025-03-28271149|StarVector is a foundation model for SVG generation that transforms vectorization into a code generation task. Using a vision-language modeling architecture, StarVector processes both visual and textual inputs to produce high-quality SVG code with remarkable precision.| |!red-down-arrow 1021|jqnatividad/qsv !2025-03-2827092|CSVs sliced, diced & analyzed.| | 1022|FranxYao/chain-of-thought-hub !2025-03-2826991|Benchmarking large language models' complex reasoning ability with chain-of-thought prompting| | 1023|princeton-nlp/SWE-bench !2025-03-2826965|[ICLR 2024] SWE-Bench: Can Language Models Resolve Real-world Github Issues?| | 1024|elastic/otel-profiling-agent !2025-03-2826930|The production-scale datacenter profiler| | 1025|src-d/hercules !2025-03-2826900|Gaining advanced insights from Git repository history.| | 1026|lanqian528/chat2api !2025-03-2826695|A service that can convert ChatGPT on the web to OpenAI API format.| | 1027|ishan0102/vimGPT !2025-03-2826681|Browse the web with GPT-4V and Vimium| | 1028|TMElyralab/MuseV !2025-03-2826650|MuseV: Infinite-length and High Fidelity Virtual Human Video Generation with Visual Conditioned Parallel Denoising| | 1029|georgia-tech-db/eva !2025-03-2826600 |AI-Relational Database System | | 1030|kubernetes-sigs/controller-runtime !2025-03-2826590|Repo for the controller-runtime subproject of kubebuilder (sig-apimachinery)| | 1031|gptlink/gptlink !2025-03-2826550 |Build your own free commercial ChatGPT environment in 10 minutes. The setup is simple and includes features such as user management, orders, tasks, and payments| | 1032|pytorch/executorch !2025-03-2826534|On-device AI across mobile, embedded and edge for PyTorch| | 1033|NVIDIA/nv-ingest !2025-03-2826290|NVIDIA Ingest is an early access set of microservices for parsing hundreds of thousands of complex, messy unstructured PDFs and other enterprise documents into metadata and text to embed into retrieval systems.| | 1034|SuperTux/supertux !2025-03-2826081|SuperTux source code| | 1035|abi/secret-llama !2025-03-2826050|Fully private LLM chatbot that runs entirely with a browser with no server needed. Supports Mistral and LLama 3.| | 1036|liou666/polyglot !2025-03-2825841 |Desktop AI Language Practice Application| | 1037|janhq/nitro !2025-03-2825821|A fast, lightweight, embeddable inference engine to supercharge your apps with local AI. OpenAI-compatible API| | 1038|deepseek-ai/DeepSeek-Math !2025-03-2825825|DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models| | 1039|anthropics/prompt-eng-interactive-tutorial !2025-03-2825781|Anthropic's Interactive Prompt Engineering Tutorial| | 1040|microsoft/promptbench !2025-03-2825741|A unified evaluation framework for large language models| | 1041|baaivision/Painter !2025-03-2825580 |Painter & SegGPT Series: Vision Foundation Models from BAAI| | 1042|OpenPipe/OpenPipe !2025-03-2825581|Turn expensive prompts into cheap fine-tuned models| | 1043|TracecatHQ/tracecat !2025-03-2825531|😼 The AI-native, open source alternative to Tines / Splunk SOAR.| | 1044|JoshuaC215/agent-service-toolkit !2025-03-2825528|Full toolkit for running an AI agent service built with LangGraph, FastAPI and Streamlit| | 1045|databricks/dbrx !2025-03-2825460|Code examples and resources for DBRX, a large language model developed by Databricks| | 1046|lamini-ai/lamini !2025-03-2825271 |Official repo for Lamini's data generator for generating instructions to train instruction-following LLMs| | 1047|mshumer/gpt-author !2025-03-282510-1|| | 1048|TMElyralab/MusePose !2025-03-2824971|MusePose: a Pose-Driven Image-to-Video Framework for Virtual Human Generation| | 1049|Kludex/fastapi-tips !2025-03-2824974|FastAPI Tips by The FastAPI Expert!| | 1050|openai/simple-evals !2025-03-2824813|| | 1051|iterative/datachain !2025-03-2824732|AI-data warehouse to enrich, transform and analyze data from cloud storages| | 1052|girafe-ai/ml-course !2025-03-2824703|Open Machine Learning course| | 1053|kevmo314/magic-copy !2025-03-2824620 |Magic Copy is a Chrome extension that uses Meta's Segment Anything Model to extract a foreground object from an image and copy it to the clipboard.| | 1054|Eladlev/AutoPrompt !2025-03-2824432|A framework for prompt tuning using Intent-based Prompt Calibration| | 1055|OpenBMB/CPM-Bee !2025-03-282434-1 |A bilingual large-scale model with trillions of parameters| | 1056|IDEA-Research/T-Rex !2025-03-2824310|T-Rex2: Towards Generic Object Detection via Text-Visual Prompt Synergy| | 1057|microsoft/genaiscript !2025-03-2824202|Automatable GenAI Scripting| | 1058|paulpierre/RasaGPT !2025-03-2824090 |💬 RasaGPT is the first headless LLM chatbot platform built on top of Rasa and Langchain. Built w/ Rasa, FastAPI, Langchain, LlamaIndex, SQLModel, pgvector, ngrok, telegram| | 1059|ashishpatel26/LLM-Finetuning !2025-03-2823911|LLM Finetuning with peft| | 1060|SoraWebui/SoraWebui !2025-03-2823570|SoraWebui is an open-source Sora web client, enabling users to easily create videos from text with OpenAI's Sora model.| | 1061|6drf21e/ChatTTScolab !2025-03-2823491|🚀 一键部署（含离线整合包）！基于 ChatTTS ，支持音色抽卡、长音频生成和分角色朗读。简单易用，无需复杂安装。| | 1062|Azure/PyRIT !2025-03-2823343|The Python Risk Identification Tool for generative AI (PyRIT) is an open access automation framework to empower security professionals and machine learning engineers to proactively find risks in their generative AI systems.| | 1063|tencent-ailab/V-Express !2025-03-2823201|V-Express aims to generate a talking head video under the control of a reference image, an audio, and a sequence of V-Kps images.| | 1064|THUDM/CogVLM2 !2025-03-2823170|GPT4V-level open-source multi-modal model based on Llama3-8B| | 1065|dvmazur/mixtral-offloading !2025-03-2823001|Run Mixtral-8x7B models in Colab or consumer desktops| | 1066|semanser/codel !2025-03-2822950|✨ Fully autonomous AI Agent that can perform complicated tasks and projects using terminal, browser, and editor.| | 1067|mshumer/gpt-investor !2025-03-2822590|| | 1068|aixcoder-plugin/aiXcoder-7B !2025-03-2822550|official repository of aiXcoder-7B Code Large Language Model| | 1069|Azure-Samples/graphrag-accelerator !2025-03-2822503|One-click deploy of a Knowledge Graph powered RAG (GraphRAG) in Azure| | 1070|emcf/engshell !2025-03-2821830 |An English-language shell for any OS, powered by LLMs| | 1071|hncboy/chatgpt-web-java !2025-03-2821771|ChatGPT project developed in Java, based on Spring Boot 3 and JDK 17, supports both AccessToken and ApiKey modes| | 1072|openai/consistencydecoder !2025-03-2821692|Consistency Distilled Diff VAE| | 1073|Alpha-VLLM/Lumina-T2X !2025-03-2821681|Lumina-T2X is a unified framework for Text to Any Modality Generation| | 1074|bghira/SimpleTuner !2025-03-2821612|A general fine-tuning kit geared toward Stable Diffusion 2.1, Stable Diffusion 3, DeepFloyd, and SDXL.| | 1075|JiauZhang/DragGAN !2025-03-2821530 |Implementation of DragGAN: Interactive Point-based Manipulation on the Generative Image Manifold| | 1076|cgpotts/cs224u !2025-03-2821390|Code for Stanford CS224u| | 1077|PKU-YuanGroup/MoE-LLaVA !2025-03-2821300|Mixture-of-Experts for Large Vision-Language Models| | 1078|darrenburns/elia !2025-03-2820831|A snappy, keyboard-centric terminal user interface for interacting with large language models. Chat with ChatGPT, Claude, Llama 3, Phi 3, Mistral, Gemma and more.| | 1079|ageerle/ruoyi-ai !2025-03-28207898|RuoYi AI 是一个全栈式 AI 开发平台，旨在帮助开发者快速构建和部署个性化的 AI 应用。| | 1080|NVIDIA/gpu-operator !2025-03-2820510|NVIDIA GPU Operator creates/configures/manages GPUs atop Kubernetes| | 1081|BAAI-Agents/Cradle !2025-03-2820481|The Cradle framework is a first attempt at General Computer Control (GCC). Cradle supports agents to ace any computer task by enabling strong reasoning abilities, self-improvment, and skill curation, in a standardized general environment with minimal requirements.| | 1082|microsoft/aici !2025-03-2820080|AICI: Prompts as (Wasm) Programs| | 1083|PRIS-CV/DemoFusion !2025-03-2820040|Let us democratise high-resolution generation! (arXiv 2023)| | 1084|apple/axlearn !2025-03-2820012|An Extensible Deep Learning Library| | 1085|naver/mast3r !2025-03-2819685|Grounding Image Matching in 3D with MASt3R| | 1086|liltom-eth/llama2-webui !2025-03-281958-1|Run Llama 2 locally with gradio UI on GPU or CPU from anywhere (Linux/Windows/Mac). Supporting Llama-2-7B/13B/70B with 8-bit, 4-bit. Supporting GPU inference (6 GB VRAM) and CPU inference.| | 1087|GaParmar/img2img-turbo !2025-03-2819582|One-step image-to-image with Stable Diffusion turbo: sketch2image, day2night, and more| | 1088|Niek/chatgpt-web !2025-03-2819560|ChatGPT web interface using the OpenAI API| | 1089|huggingface/cookbook !2025-03-2819421|Open-source AI cookbook| | 1090|pytorch/ao !2025-03-2819241|PyTorch native quantization and sparsity for training and inference| | 1091|emcie-co/parlant !2025-03-2819053|The behavior guidance framework for customer-facing LLM agents| | 1092|ymcui/Chinese-LLaMA-Alpaca-3 !2025-03-2818980|中文羊驼大模型三期项目 (Chinese Llama-3 LLMs) developed from Meta Llama 3| | 1093|Nutlope/notesGPT !2025-03-2818811|Record voice notes & transcribe, summarize, and get tasks| | 1094|InstantStyle/InstantStyle !2025-03-2818791|InstantStyle: Free Lunch towards Style-Preserving in Text-to-Image Generation 🔥| | 1095|idaholab/moose !2025-03-2818771|Multiphysics Object Oriented Simulation Environment| | 1096|The-OpenROAD-Project/OpenROAD !2025-03-2818351|OpenROAD's unified application implementing an RTL-to-GDS Flow. Documentation at https://openroad.readthedocs.io/en/latest/| | 1097|alibaba/spring-ai-alibaba !2025-03-281831121|Agentic AI Framework for Java Developers| | 1098|ytongbai/LVM !2025-03-2817990|Sequential Modeling Enables Scalable Learning for Large Vision Models| | 1099|microsoft/sample-app-aoai-chatGPT !2025-03-2817981|[PREVIEW] Sample code for a simple web chat experience targeting chatGPT through AOAI.| | 1100|AI-Citizen/SolidGPT !2025-03-2817830|Chat everything with your code repository, ask repository level code questions, and discuss your requirements. AI Scan and learning your code repository, provide you code repository level answer🧱 🧱| | 1101|YangLing0818/RPG-DiffusionMaster !2025-03-2817784|Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (PRG)| | 1102|kyegomez/BitNet !2025-03-2817710|Implementation of "BitNet: Scaling 1-bit Transformers for Large Language Models" in pytorch| | 1103|eloialonso/diamond !2025-03-2817671|DIAMOND (DIffusion As a Model Of eNvironment Dreams) is a reinforcement learning agent trained in a diffusion world model.| | 1104|flowdriveai/flowpilot !2025-03-2817250|flow-pilot is an openpilot based driver assistance system that runs on linux, windows and android powered machines.| | 1105|xlang-ai/OSWorld !2025-03-2817200|OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments| | 1106|linyiLYi/snake-ai !2025-03-2817031|An AI agent that beats the classic game "Snake".| | 1107|baaivision/Emu !2025-03-2816991|Emu Series: Generative Multimodal Models from BAAI| | 1108|kevmo314/scuda !2025-03-2816870|SCUDA is a GPU over IP bridge allowing GPUs on remote machines to be attached to CPU-only machines.| | 1109|SharifiZarchi/IntroductiontoMachineLearning !2025-03-2816701|دوره‌ی مقدمه‌ای بر یادگیری ماشین، برای دانشجویان| | 1110|google/maxtext !2025-03-2816670|A simple, performant and scalable Jax LLM!| | 1111|ml-explore/mlx-swift-examples !2025-03-2816471|Examples using MLX Swift| | 1112|unitreerobotics/unitreerlgym !2025-03-2816256|| | 1113|collabora/WhisperFusion !2025-03-2815901|WhisperFusion builds upon the capabilities of WhisperLive and WhisperSpeech to provide a seamless conversations with an AI.| | 1114|lichao-sun/Mora !2025-03-2815520|Mora: More like Sora for Generalist Video Generation| | 1115|GoogleCloudPlatform/localllm !2025-03-2815370|Run LLMs locally on Cloud Workstations| | 1116|TencentARC/BrushNet !2025-03-2815330|The official implementation of paper "BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion"| | 1117|ai-christianson/RA.Aid !2025-03-2815288|Develop software autonomously.| | 1118|stephansturges/WALDO !2025-03-2815170|Whereabouts Ascertainment for Low-lying Detectable Objects. The SOTA in FOSS AI for drones!| | 1119|skills/copilot-codespaces-vscode !2025-03-2815112|Develop with AI-powered code suggestions using GitHub Copilot and VS Code| | 1120|andrewnguonly/Lumos !2025-03-2814920|A RAG LLM co-pilot for browsing the web, powered by local LLMs| | 1121|TeamNewPipe/NewPipeExtractor !2025-03-2814811|NewPipe's core library for extracting data from streaming sites| | 1122|mhamilton723/FeatUp !2025-03-2814770|Official code for "FeatUp: A Model-Agnostic Frameworkfor Features at Any Resolution" ICLR 2024| | 1123|AnswerDotAI/fsdpqlora !2025-03-2814671|Training LLMs with QLoRA + FSDP| | 1124|jgravelle/AutoGroq !2025-03-2814330|| | 1125|OpenGenerativeAI/llm-colosseum !2025-03-2814130|Benchmark LLMs by fighting in Street Fighter 3! The new way to evaluate the quality of an LLM| | 1126|microsoft/vscode-ai-toolkit !2025-03-2814000|| | 1127|McGill-NLP/webllama !2025-03-2813930|Llama-3 agents that can browse the web by following instructions and talking to you| | 1128|lucidrains/self-rewarding-lm-pytorch !2025-03-2813760|Implementation of the training framework proposed in Self-Rewarding Language Model, from MetaAI| | 1129|ishaan1013/sandbox !2025-03-2813650|A cloud-based code editing environment with an AI copilot and real-time collaboration.| | 1130|goatcorp/Dalamud !2025-03-2813275|FFXIV plugin framework and API| | 1131|Lightning-AI/lightning-thunder !2025-03-2813151|Make PyTorch models Lightning fast! Thunder is a source to source compiler for PyTorch. It enables using different hardware executors at once.| | 1132|PKU-YuanGroup/MagicTime !2025-03-2813052|MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators| | 1133|SakanaAI/evolutionary-model-merge !2025-03-2813000|Official repository of Evolutionary Optimization of Model Merging Recipes| | 1134|a-real-ai/pywinassistant !2025-03-2812950|The first open source Large Action Model generalist Artificial Narrow Intelligence that controls completely human user interfaces by only using natural language. PyWinAssistant utilizes Visualization-of-Thought Elicits Spatial Reasoning in Large Language Models.| | 1135|TraceMachina/nativelink !2025-03-2812630|NativeLink is an open source high-performance build cache and remote execution server, compatible with Bazel, Buck2, Reclient, and other RBE-compatible build systems. It offers drastically faster builds, reduced test flakiness, and significant infrastructure cost savings.| | 1136|MLSysOps/MLE-agent !2025-03-2812500|🤖 MLE-Agent: Your intelligent companion for seamless AI engineering and research. 🔍 Integrate with arxiv and paper with code to provide better code/research plans 🧰 OpenAI, Ollama, etc supported. 🎆 Code RAG| | 1137|wpilibsuite/allwpilib !2025-03-2811610|Official Repository of WPILibJ and WPILibC| | 1138|elfvingralf/macOSpilot-ai-assistant !2025-03-2811470|Voice + Vision powered AI assistant that answers questions about any application, in context and in audio.| | 1139|langchain-ai/langchain-extract !2025-03-2811210|🦜⛏️ Did you say you like data?| | 1140|FoundationVision/GLEE !2025-03-2811120|【CVPR2024】GLEE: General Object Foundation Model for Images and Videos at Scale| | 1141|Profluent-AI/OpenCRISPR !2025-03-2810990|AI-generated gene editing systems| | 1142|zju3dv/EasyVolcap !2025-03-2810821|[SIGGRAPH Asia 2023 (Technical Communications)] EasyVolcap: Accelerating Neural Volumetric Video Research| | 1143|PaddlePaddle/PaddleHelix !2025-03-2810560|Bio-Computing Platform Featuring Large-Scale Representation Learning and Multi-Task Deep Learning “螺旋桨”生物计算工具集| | 1144|myshell-ai/JetMoE !2025-03-289800|Reaching LLaMA2 Performance with 0.1M Dollars| | 1145|likejazz/llama3.np !2025-03-289770|llama3.np is pure NumPy implementation for Llama 3 model.| | 1146|mustafaaljadery/gemma-2B-10M !2025-03-289500|Gemma 2B with 10M context length using Infini-attention.| | 1147|HITsz-TMG/FilmAgent !2025-03-289382|Resources of our paper "FilmAgent: A Multi-Agent Framework for End-to-End Film Automation in Virtual 3D Spaces". New versions in the making!| | 1148|aws-samples/amazon-bedrock-samples !2025-03-289362|This repository contains examples for customers to get started using the Amazon Bedrock Service. This contains examples for all available foundational models| | 1149|Akkudoktor-EOS/EOS !2025-03-2893154|This repository features an Energy Optimization System (EOS) that optimizes energy distribution, usage for batteries, heat pumps& household devices. It includes predictive models for electricity prices (planned), load forecasting& dynamic optimization to maximize energy efficiency & minimize costs. Founder Dr. Andreas Schmitz (YouTube @akkudoktor)| Tip: | symbol| rule | | :----| :---- | |🔥 | 256 1k| |!green-up-arrow.svg !red-down-arrow | ranking up / down| |⭐ | on trending page today| [Back to Top] Tools | No. | Tool | Description | | ----:|:----------------------------------------------- |:------------------------------------------------------------------------------------------- | | 1 | ChatGPT | A sibling model to InstructGPT, which is trained to follow instructions in a prompt and provide a detailed response | | 2 | DALL·E 2 | Create original, realistic images and art from a text description | | 3 | Murf AI | AI enabled, real people's voices| | 4 | Midjourney | An independent research lab that produces an artificial intelligence program under the same name that creates images from textual descriptions, used in Discord | 5 | Make-A-Video | Make-A-Video is a state-of-the-art AI system that generates videos from text | | 6 | Creative Reality™ Studio by D-ID| Use generative AI to create future-facing videos| | 7 | chat.D-ID| The First App Enabling Face-to-Face Conversations with ChatGPT| | 8 | Notion AI| Access the limitless power of AI, right inside Notion. Work faster. Write better. Think bigger. | | 9 | Runway| Text to Video with Gen-2 | | 10 | Resemble AI| Resemble’s AI voice generator lets you create human–like voice overs in seconds | | 11 | Cursor| Write, edit, and chat about your code with a powerful AI | | 12 | Hugging Face| Build, train and deploy state of the art models powered by the reference open source in machine learning | | 13 | Claude | A next-generation AI assistant for your tasks, no matter the scale | | 14 | Poe| Poe lets you ask questions, get instant answers, and have back-and-forth conversations with AI. Gives access to GPT-4, gpt-3.5-turbo, Claude from Anthropic, and a variety of other bots| [Back to Top] Websites | No. | WebSite |Description | | ----:|:------------------------------------------ |:---------------------------------------------------------------------------------------- | | 1 | OpenAI | An artificial intelligence research lab | | 2 | Bard | Base Google's LaMDA chatbots and pull from internet | | 3 | ERNIE Bot | Baidu’s new generation knowledge-enhanced large language model is a new member of the Wenxin large model family | | 4 | DALL·E 2 | An AI system that can create realistic images and art from a description in natural language | | 5 | Whisper | A general-purpose speech recognition model | | 6| CivitAI| A platform that makes it easy for people to share and discover resources for creating AI art| | 7|D-ID| D-ID’s Generative AI enables users to transform any picture or video into extraordinary experiences| | 8| Nvidia eDiff-I| Text-to-Image Diffusion Models with Ensemble of Expert Denoisers | | 9| Stability AI| The world's leading open source generative AI company which opened source Stable Diffusion | | 10| Meta AI| Whether it be research, product or infrastructure development, we’re driven to innovate responsibly with AI to benefit the world | | 11| ANTHROPIC| AI research and products that put safety at the frontier | [Back to Top] Reports&Papers | No. | Report&Paper | Description | |:---- |:-------------------------------------------------------------------------------------------------------------- |:---------------------------------------------------- | | 1 | GPT-4 Technical Report | GPT-4 Technical Report | | 2 | mli/paper-reading | Deep learning classics and new papers are read carefully paragraph by paragraph. | | 3 | labmlai/annotateddeeplearningpaperimplementations| A collection of simple PyTorch implementations of neural networks and related algorithms, which are documented with explanations | | 4 | Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models | Talking, Drawing and Editing with Visual Foundation Models | | 5 | OpenAI Research | The latest research report and papers from OpenAI | | 6 | Make-A-Video: Text-to-Video Generation without Text-Video Data|Meta's Text-to-Video Generation| | 7 | eDiff-I: Text-to-Image Diffusion Models with Ensemble of Expert Denoisers| Nvidia eDiff-I - New generation of generative AI content creation tool | | 8 | Training an Assistant-style Chatbot with Large Scale Data Distillation from GPT-3.5-Turbo | 2023 GPT4All Technical Report | | 9 | Segment Anything| Meta Segment Anything | | 10 | LLaMA: Open and Efficient Foundation Language Models| LLaMA: a collection of foundation language models ranging from 7B to 65B parameters| | 11 | papers-we-love/papers-we-love |Papers from the computer science community to read and discuss| | 12 | CVPR 2023 papers |The most exciting and influential CVPR 2023 papers| [Back to Top] Tutorials | No. | Tutorial | Description| |:---- |:---------------------------------------------------------------- | --- | | 1 | Coursera - Machine Learning | The Machine Learning Specialization Course taught by Dr. Andrew Ng| | 2 | microsoft/ML-For-Beginners | 12 weeks, 26 lessons, 52 quizzes, classic Machine Learning for all| | 3 | ChatGPT Prompt Engineering for Developers | This short course taught by Isa Fulford (OpenAI) and Andrew Ng (DeepLearning.AI) will teach how to use a large language model (LLM) to quickly build new and powerful applications | | 4 | Dive into Deep Learning |Targeting Chinese readers, functional and open for discussion. The Chinese and English versions are used for teaching in over 400 universities across more than 60 countries | | 5 | AI Expert Roadmap | Roadmap to becoming an Artificial Intelligence Expert in 2022 | | 6 | Computer Science courses |List of Computer Science courses with video lectures| | 7 | Machine Learning with Python | Machine Learning with Python Certification on freeCodeCamp| | 8 | Building Systems with the ChatGPT API | This short course taught by Isa Fulford (OpenAI) and Andrew Ng (DeepLearning.AI), you will learn how to automate complex workflows using chain calls to a large language model| | 9 | LangChain for LLM Application Development | This short course taught by Harrison Chase (Co-Founder and CEO at LangChain) and Andrew Ng. you will gain essential skills in expanding the use cases and capabilities of language models in application development using the LangChain framework| | 10 | How Diffusion Models Work | This short course taught by Sharon Zhou (CEO, Co-founder, Lamini). you will gain a deep familiarity with the diffusion process and the models which carry it out. More than simply pulling in a pre-built model or using an API, this course will teach you to build a diffusion model from scratch| | 11 | Free Programming Books For AI |📚 Freely available programming books for AI | | 12 | microsoft/AI-For-Beginners |12 Weeks, 24 Lessons, AI for All!| | 13 | hemansnation/God-Level-Data-Science-ML-Full-Stack |A collection of scientific methods, processes, algorithms, and systems to build stories & models. This roadmap contains 16 Chapters, whether you are a fresher in the field or an experienced professional who wants to transition into Data Science & AI| | 14 | datawhalechina/prompt-engineering-for-developers |Chinese version of Andrew Ng's Big Model Series Courses, including "Prompt Engineering", "Building System", and "LangChain"| | 15 | ossu/computer-science |🎓 Path to a free self-taught education in Computer Science!| | 16 | microsoft/Data-Science-For-Beginners | 10 Weeks, 20 Lessons, Data Science for All! | |17 |jwasham/coding-interview-university !2023-09-29268215336 |A complete computer science study plan to become a software engineer.| [Back to Top] Thanks If this project has been helpful to you in any way, please give it a ⭐️ by clicking on the star.

#github #aitreasurebox

github

0.535

0.022778788676674117

trypromptly•Mar 28, 2025

LLMStack

LLMStack is a no-code platform for building generative AI agents, workflows and chatbots, connecting them to your data and business processes. Quickstart | Documentation | Promptly Overview Build tailor-made generative AI agents, applications and chatbots that cater to your unique needs by chaining multiple LLMs. Seamlessly integrate your own data, internal tools and GPT-powered models without any coding experience using LLMStack's no-code builder. Trigger your AI chains from Slack or Discord. Deploy to the cloud or on-premise. !llmstack-quickstart See full demo video here Getting Started Check out our Cloud offering at Promptly or follow the instructions below to deploy LLMStack on your own infrastructure. LLMStack deployment comes with a default admin account whose credentials are admin and promptly. Be sure to change the password from admin panel after logging in. Installation Prerequisites LLMStack depends on a background docker container to run jobs. Make sure you have Docker installed on your machine if want to use jobs. You can follow the instructions here to install Docker. Install LLMStack using pip If you are on windows, please use WSL2 (Windows Subsystem for Linux) to install LLMStack. You can follow the instructions here to install WSL2. Once you are in a WSL2 terminal, you can install LLMStack using the above command. Start LLMStack using the following command: Above commands will install and start LLMStack. It will create .llmstack in your home directory and places the database and config files in it when run for the first time. Once LLMStack is up and running, it should automatically open your browser and point it to localhost:3000. You can add your own keys to providers like OpenAI, Cohere, Stability etc., from Settings page. If you want to provide default keys for all the users of your LLMStack instance, you can add them to the ~/.llmstack/config file. LLMStack: Quickstart video Features 🤖 Agents: Build generative AI agents like AI SDRs, Research Analysts, RPA Automations etc., without writing any code. Connect agents to your internal or external tools, search the web or browse the internet with agents. 🔗 Chain multiple models: LLMStack allows you to chain multiple LLMs together to build complex generative AI applications. 📊 Use generative AI on your Data: Import your data into your accounts and use it in AI chains. LLMStack allows importing various types (CSV, TXT, PDF, DOCX, PPTX etc.,) of data from a variety of sources (gdrive, notion, websites, direct uploads etc.,). Platform will take care of preprocessing and vectorization of your data and store it in the vector database that is provided out of the box. 🛠️ No-code builder: LLMStack comes with a no-code builder that allows you to build AI chains without any coding experience. You can chain multiple LLMs together and connect them to your data and business processes. ☁️ Deploy to the cloud or on-premise: LLMStack can be deployed to the cloud or on-premise. You can deploy it to your own infrastructure or use our cloud offering at Promptly. 🚀 API access: Apps or chatbots built with LLMStack can be accessed via HTTP API. You can also trigger your AI chains from Slack or Discord. 🏢 Multi-tenant: LLMStack is multi-tenant. You can create multiple organizations and add users to them. Users can only access the data and AI chains that belong to their organization. What can you build with LLMStack? Using LLMStack you can build a variety of generative AI applications, chatbots and agents. Here are some examples: 👩🏻‍💼 AI SDRs: You can build AI SDRs (Sales Development Representatives) that can generate personalized emails, LinkedIn messages, cold calls, etc., for your sales team 👩🏻‍💻 Research Analysts: You can build AI Research Analysts that can generate research reports, investment thesis, etc., for your investment team 🤖 RPA Automations: You can build RPA automations that can automate your business processes by generating emails, filling forms, etc., 📝 Text generation: You can build apps that generate product descriptions, blog posts, news articles, tweets, emails, chat messages, etc., by using text generation models and optionally connecting your data. Check out this marketing content generator for example 🤖 Chatbots: You can build chatbots trained on your data powered by ChatGPT like Promptly Help that is embedded on Promptly website 🎨 Multimedia generation: Build complex applications that can generate text, images, videos, audio, etc. from a prompt. This story generator is an example 🗣️ Conversational AI: Build conversational AI systems that can have a conversation with a user. Check out this Harry Potter character chatbot 🔍 Search augmentation: Build search augmentation systems that can augment search results with additional information using APIs. Sharebird uses LLMStack to augment search results with AI generated answer from their content similar to Bing's chatbot 💬 Discord and Slack bots: Apps built on LLMStack can be triggered from Slack or Discord. You can easily connect your AI chains to Slack or Discord from LLMStack's no-code app editor. Check out our Discord server to interact with one such bot. Administration Login to http://localhost:3000/admin using the admin account. You can add users and assign them to organizations in the admin panel. Cloud Offering Check out our cloud offering at Promptly. You can sign up for a free account and start building your own generative AI applications. Documentation Check out our documentation at docs.trypromptly.com/llmstack to learn more about LLMStack. Development Check out our development guide at docs.trypromptly.com/llmstack/development to learn more about how to run and develop LLMStack. Contributing We welcome contributions to LLMStack. Please check out our contributing guide to learn more about how you can contribute to LLMStack.

ai-flow

Open-source tool to seamlessly connect multiple AI model APIs into repeatable workflows. 🔗 Website • 📚 Documentation 🎉🚀 Latest Release: v0.10.0 🚀🎉 New Nodes: Claude 3.7, OpenRouter, Generate Random Number Configuration can now be done entirely in the UI !AI-Flow Intro Overview AI-Flow is an open-source, user-friendly UI that lets you visually design, manage, and monitor AI-driven workflows by seamlessly connecting multiple AI model APIs (e.g., OpenAI, StabilityAI, Replicate, Claude, Deepseek). Features Visual Workflow Builder: Drag-and-drop interface for crafting AI workflows. Real-Time Monitoring: Watch your workflow execute and track results. Parallel Processing: Nodes run in parallel whenever possible. Model Management: Easily organize and manage diverse AI models. Import/Export: Share or back up your workflows effortlessly. Supported Models Replicate: LLaMa, Mistral, FaceSwap, InstantMesh, MusicGen, and more. OpenAI: GPT-4o, TTS, o1, o3. StabilityAI: Stable Diffusion 3.5, SDXL, Stable Video Diffusion, plus additional tools. Others: Claude, Deepseek. !Scenario Example Open Source vs. Cloud AI-Flow is fully open source and available under the MIT License, empowering you to build and run your AI workflows on your personal machine. For those seeking enhanced functionality and a polished experience, AI-Flow Pro on our cloud platform (app.ai-flow.net) offers advanced features, including: Subflows & Loops: Create complex, nested workflows and iterate tasks effortlessly. API-Triggered Flows: Initiate workflows via API calls for seamless automation. Integrated Services: Connect with external services such as Google Search, Airtable, Zapier, and Make. Simplified Interface: Transform workflows into streamlined tools with an intuitive UI. !Pro VS Open Source The cloud version builds upon the foundation of the open-source project, giving you more power and flexibility while still letting you use your own API keys. Installation Note: To unlock full functionality, AI-Flow requires S3-compatible storage (with proper CORS settings) to host resources. Without it, features like File Upload or nodes that rely on external providers (e.g., StabilityAI) may not work as expected. Also, set REPLICATEAPIKEY in your environment to use the Replicate node. Local Installation (Without Docker) Clone the Repository: UI Setup: Backend Setup: Windows Users: Run the Application: Start the backend: In a new terminal, start the UI: Open your browser and navigate to http://localhost:3000. Docker Installation Prepare Docker Compose: Navigate to the docker directory: Update the REPLICATEAPIKEY in the YAML file. Launch with Docker Compose: Access the Application: Open http://localhost:80 in your browser. To stop, run: Contributing We welcome contributions! If you encounter issues or have feature ideas, please open an issue or submit a pull request. License This project is released under the MIT License.

video-killed-the-radio-star

Video Killed The Radio Star Requirements ffmpeg - https://ffmpeg.org/ pytorch - https://pytorch.org/get-started/locally/ vktrs - (this repo) - pip install vktrs[api] stability_sdk api token - https://beta.dreamstudio.ai/ > circular icon in top right > membership > API Key whisper - pip install git+https://github.com/openai/whisper FAQ What is this? TLDR: Automated music video maker, given an mp3 or a youtube URL How does this animation technique work? For each text prompt you provide, the notebook will... Generate an image based on that text prompt (using stable diffusion) Use the generated image as the init_image to recombine with the text prompt to generate variations similar to the first image. This produces a sequence of extremely similar images based on the original text prompt Images are then intelligently reordered to find the smoothest animation sequence of those frames This image sequence is then repeated to pad out the animation duration as needed The technique demonstrated in this notebook was inspired by a video created by Ben Gillin. How are lyrics transcribed? This notebook uses openai's recently released 'whisper' model for performing automatic speech recognition. OpenAI was kind of to offer several different sizes of this model which each have their own pros and cons. This notebook uses the largest whisper model for transcribing the actual lyrics. Additionally, we use the smallest model for performing the lyric segmentation. Neither of these models is perfect, but the results so far seem pretty decent. The first draft of this notebook relied on subtitles from youtube videos to determine timing, which was then aligned with user-provided lyrics. Youtube's automated captions are powerful and I'll update the notebook shortly to leverage those again, but for the time being we're just using whisper for everything and not referencing user-provided captions at all. Something didn't work quite right in the transcription process. How do fix the timing or the actual lyrics? The notebook is divided into several steps. Between each step, a "storyboard" file is updated. If you want to make modifications, you can edit this file directly and those edits should be reflected when you next load the file. Depending on what you changed and what step you run next, your changes may be ignored or even overwritten. Still playing with different solutions here. Can I provide my own images to 'bring to life' and associate with certain lyrics/sequences? Yes, you can! As described above: you just need to modify the storyboard. Will describe this functionality in greater detail after the implementation stabilizes a bit more. This gave me an idea and I'd like to use just a part of your process here. What's the best way to reuse just some of the machinery you've developed here? Most of the functionality in this notebook has been offloaded to library I published to pypi called vktrs. I strongly encourage you to import anything you need from there rather than cutting and pasting function into a notebook. Similarly, if you have ideas for improvements, please don't hesitate to submit a PR! Dev notes

#github #video-killed-the-radio-star

AI News: Vibe Jam, The BEST Small LLM, Claude Search, OpenAI Audio Models, and more!

youtube

0.344

0.57

Matthew Berman•Mar 21, 2025

AI News: Vibe Jam, The BEST Small LLM, Claude Search, OpenAI Audio Models, and more!

AI News is back! Let me know down below if you want me to start making weekly news videos again! Join My Newsletter for Regular AI Updates 👇🏼 https://forwardfuture.ai My Links 🔗 👉🏻 Subscribe: https://www.youtube.com/@matthew_berman 👉🏻 Twitter: https://twitter.com/matthewberman 👉🏻 Discord: https://discord.gg/xxysSXBxFW 👉🏻 Patreon: https://patreon.com/MatthewBerman 👉🏻 Instagram: https://www.instagram.com/matthewberman_ai 👉🏻 Threads: https://www.threads.net/@matthewberman_ai 👉🏻 LinkedIn: https://www.linkedin.com/company/forward-future-ai Media/Sponsorship Inquiries ✅ https://bit.ly/44TC45V Timestamps: 0:00 - Vibe Coding Jam 1:20 - Mistral Model 2:20 - Claude Search 2:57 - OpenAI Audio 4:39 - Windsurf Wave 5 5:28 - KREA AI Video Training 5:52 - Notebook LM Mindmaps 6:13 - Hunyuan 3D Modeling AI 6:45 - Stability AI Stable Virtual Camera 7:22 - Gemini Canvas 7:41 - LG EXAONE Links: https://x.com/levelsio/status/1901660771505021314 https://x.com/bgdnandrew/status/1903093702827933879 https://x.com/fergarram/status/1902779314224996624 https://x.com/LBacaj/status/1903114707336048851 https://x.com/realitybndr/status/1903054190944960824 https://x.com/dev_clap/status/1902912215860060494 https://x.com/iamarkdev/status/1902131862421836120 https://x.com/gabebusto/status/1902727892284379324 https://x.com/scobelverse/status/1903069102777962554/photo/1 https://mistral.ai/news/mistral-small-3-1 https://x.com/MistralAI/status/1901668499832918151 https://x.com/AnthropicAI/status/1902765011727999046 https://x.com/OpenAIDevs/status/1902773659497885936 https://www.openai.fm/ https://openai.com/index/introducing-our-next-generation-audio-models/ https://x.com/TXhunyuan/status/1901831819655606598 https://stability.ai/news/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control https://x.com/LGAIResearch/status/1901803002052436323