French AI startup Mistral is introducing new AI model customization options, including paid plans, to enable developers and enterprises to tune generative models for specific use cases.
The first one is self-service. Mistral has released a software development kit (SDK), Mistral Finetuneto tune their models on workstations, servers and small data center nodes.
In the Readme file of the SDK repository on GitHub, Mistral notes that the SDK is optimized for multi-GPU setups, but will be scaled down to a single Nvidia A100 or H100 GPU for tuning smaller models like the Mistral 7B. Tuning a dataset like UltraChat, a collection of 1.4 million dialogs from OpenAI’s ChatGPT, takes about half an hour using Mistral-Finetune on eight H100s, Mistral says.
For developers and firms that prefer a more managed solution, the newly launched Mistral tuning services are available via the company’s API. Compatible with two Mistral models for now, the Mistral Small and the aforementioned Mistral 7B, Mistral says the tuning services will gain support for more models in the coming weeks.
Finally, Mistral is introducing custom training services, currently only available to select customers, designed to fine-tune any Mistral model for an organization’s applications using their data. “This approach enables the creation of highly specialized and optimized models for their specific domain,” the company explains in a post on its official profile blog.
Mistral, which was recently reported by my colleague Ingrid Lunden, is in search of to raise roughly $600 million at a $6 billion valuation from investors including DST, General Catalyst and Lightspeed Venture Partners, is little doubt eager to increase revenue in the face of serious – and growing – competition in the generative AI space market.
Since Mistral unveiled its first generative model in September 2023, it has released several more, including code generation modeland we have implemented paid APIs. However, he didn’t disclose how many users he has or what his revenues are.