Replicate’s library of over 25,000 open source templates has been used by over 2 million software developers. With the new funding, the co-founders want to expand their user base.
Reproduce
Amid the AI land grab, open source models whose code can be modified and used by anyone are catching up with their proprietary counterparts. They are getting bigger and bigger (Meta’s Llama 2 is trained on 70 billion parameters) and are even becoming better than ChatGPT at perform specific tasks. Today, as interest in open source AI tools increases, it’s a great time to be a startup that hosts and manages open source software, said Ben Firshman, CEO of Replicate, a platform form used by 2 million software developers to access and tinker with more than 25,000 open source AI models.
When popular open source models such as the Stable Diffusion 2.0 text-image model and Meta’s large Llama 2 language model were added to its library this year, Replicate saw its biggest growth spurts. The platform also saw an uptick in traffic after OpenAI’s leadership underwent a sudden but temporary shakeup in late November.
“People are becoming more interested in open source models because they don’t want to be locked into a proprietary platform that might disappear at some point,” Firshman said.
The startup announced Tuesday that it raised $40 million in Series B funding in a round led by Andreessen Horowitz with participation from Nvidia’s VC arm, NVentures, Heavybit, Sequoia and Y Combinator. The round values the San Francisco-based startup at $350 million, according to sources familiar with the matter, and brings Replicate’s total funding to around $58 million.
Firshman and co-founder and CTO Andreas Jansson discovered the need for a platform that would make it easier for software developers to use the latest AI models ahead of today’s largest open source models, like Llama 2 and Stable Diffusion of Meta, do not enter the mainstream.
While working as a machine learning engineer at Spotify, Jansson realized that most advances in AI were locked in academic research, hidden behind long descriptions and complicated diagrams, making them useless for solve real-world problems. In 2019, he teamed up with his former colleague Firshman, who had created a system for developers to package and ship their work while running the product at software unicorn Docker, to launch Replicate. Their goal was to do the same for researchers by making their open source AI and machine learning software available to others.
“People are becoming more interested in open source models because they don’t want to be locked into a proprietary platform that might disappear at some point. »
Replicate isn’t the only startup providing compute resources to run open source models. Competition comes from both high-value startups like Together AI, which recently raised $102.5 million Series A, $4.5 billion Hugging Face and OctoML, valued at $850 million, as well as tech giants like Nvidia, Google, Amazon and Microsoft, all of which offer similar products to run and customize machine learning models on the cloud.
One argument against open source models is that there are a series of security risks: they can be used to malicious purposes like technical phishing and biological attacks, but proponents of open source argue that the fact that the model’s code is transparent means it will face greater scrutiny, which ultimately , will make the models safer. Replicate has partly solved this problem with filters that detect and prevent models from generating harmful content. But because gatekeepers tend to wrongly flag safe content as unsafe, they can be disabled.
Yet a growing demand for open source templates is evident on the Replicate platform, which hosts templates capable of generating and editing music, video, text and images. A face restoration“The AI model that can convert old, blurry photos into sharp images has been used around 60 million times. Another AI model that can to exchange a face to another face in two seconds has been performed almost 30 million times. This is partly because large open source AI models can be fine-tuned for specific use cases through training on custom data, making them cheaper and faster to use, a said Firstshman. “You can tweak some of these models for a dollar and 10 minutes,” he said.
But this fine tuning is also a more complicated process than the models you can use commercially. “Open source technology is harder to use than closed products, almost by definition,” said Matt Bornstein, a partner at Andreessen Horowitz who led the round.
Reproduce charges developers for the period of running a model, ranging from 36 cents to $20 per hour. The startup has partnered with NVIDIA to provide GPUs of various sizes and capacities and works with several cloud providers like Coreweave and Google Cloud. “Compared to a lot of other AI companies, we have a very clear business model in the sense that we sell infrastructure in exchange for money,” he said, adding that the startup was not still profitable.
The new funding will be used to attract more software developers to the platform and provide enterprise customers with additional services such as security, compliance and monitoring a model’s performance. Its 30,000 paying customers include companies like Buzzfeed, Getty-owned Unsplash, and startups like Character AI and Labelbox that use Replicate to run open source models. A software engineer himself who has already founded three tech startups, Firshman admits that he is still not a machine learning expert like his co-founder Jansson. But thanks to Replicate, which simplifies the use of open source AI models, the technology is more accessible to him and others like him.
“Machine learning models can do a lot of the boring bug-killing work that software developers actually spend most of their time doing,” he said. “I can do fun creative things.”