Almost everyone these days is trying to get a piece of generative AI work. While most of the focus remains on model vendors like OpenAI, Anthropic, and Cohere, or large corporations like Microsoft, Meta, Google, and Amazon, there are actually many startups trying to attack the generative AI problem in a variety of ways. method.
Fireworks.ai is one such startup. Although it lacks brand awareness compared to some other players, it boasts the largest open source model API, with over 12,000 users per company. This kind of open source traction tends to attract investor attention, and the company has raised $25 million to date.
Lin Qiao, co-founder and CEO of Fireworks, points out that his company is not teaching a foundational model from scratch, but rather helping to fine-tune other models to fit the specific needs of the business. “It could be an off-the-shelf model, an open source model, a model that we adapt, or a model that customers can adapt themselves,” he said. All three varieties can be provided through the inference engine API,” Qiao told TechCrunch.
Because it’s an API, developers can connect it to their applications, get selection models trained on their data, and add generative AI features like asking questions very quickly. Qiao says this is fast, efficient and produces high-quality results.
Another advantage of the Firework approach is that it allows companies to experiment with multiple models, which is important in a rapidly changing market. “Our philosophy is to help users iterate and experiment with multiple models and provide effective tools to inject data into multiple models and test products,” she said.
Perhaps more importantly, it reduces costs by limiting model size to 7 to 13 billion tokens, compared to over 1 trillion tokens in ChatGPT4. This limits the range of words that large language models can understand, but allows developers to focus on much smaller, more focused datasets designed to work on more limited business use cases.
Qiao’s previous experience working at Meta makes him uniquely qualified to build such systems, and he leads the AI platform development team with the goal of building a fast and scalable development engine that will power AI across all of Meta’s products and services. . Her work at Meta gave her this knowledge and allowed her to create API-based tools that provide this kind of functionality to any company without requiring the level of engineering resources of a company the size of Meta.
The company raised $25 million in 2022, led by Benchmark, with participation from Sequoia Capital and angel investors including Databricks and Snowflake. The latter is a particularly interesting strategic investor because both are data storage tools, and Fireworks allows users to put that data to work.