French AI startup Mistral is introducing new AI model customization options, including paid plans, to enable developers and enterprises to fine-tune generative models for specific use cases.
The first is self-service: Mistral has released a software development kit (SDK). Mistral FinetuneFine-tune your models on workstations, servers, and small datacenter nodes.
In the Readme on the SDK’s GitHub repository, Mistral says that the SDK is optimized for multi-GPU setups, but can be scaled down to a single Nvidia A100 or H100 GPU to fine-tune smaller models like the Mistral 7B. Fine-tuning a dataset such as UltraChat, a collection of 1.4 million dialogues using OpenAI’s ChatGPT, takes about 30 minutes using Mistral-Finetune on eight H100s, Mistral says.
For developers and businesses who prefer a more managed solution, there’s Mistral’s newly launched Tweak Service, available through the company’s API. Currently it’s compatible with two Mistral models: Mistral Small and the aforementioned Mistral 7B, but Mistral says the Tweak Service will be supported on more models in the coming weeks.
Finally, Mistral is launching a custom training service, currently available to select customers, that uses your data to fine-tune Mistral models for your apps. “This approach allows us to create highly specialized and optimized models for specific domains,” the company explains in a blog post. blog.
As my colleague Ingrid Lunden recently reported, Mistral is seeking to raise around $600 million from investors including DST, General Catalyst and Lightspeed Venture Partners at a $6 billion valuation, no doubt looking to grow revenue as it faces stiff (and growing) competition in the generative AI space.
Mistral announced its first generative model in September 2023 and has since released several more models, including Code Generation ModelThe company has rolled out a paid API, but has not disclosed how many users it expects to have or how much revenue it will generate.