French AI startup Mistral is expanding its offerings by introducing new options for customizing AI models. These include a range of paid plans designed to help developers and enterprises tailor Mistral’s generative models to their specific needs.
Self-Service Fine-Tuning with Mistral-Finetune SDK
One of the new offerings is a self-service option. Mistral has launched a software development kit (SDK) called Mistral-Finetune, which allows users to fine-tune the company’s models on various hardware setups, from workstations to servers and small data center nodes.
According to the SDK’s GitHub repository readme, Mistral-Finetune is optimized for multi-GPU configurations but can also run on a single Nvidia A100 or H100 GPU for tuning smaller models like Mistral 7B. For instance, fine-tuning a dataset such as UltraChat, which includes 1.4 million dialogues with OpenAI’s ChatGPT, takes about half an hour using eight H100 GPUs.
For those preferring a more hands-off approach, Mistral has introduced managed fine-tuning services accessible through the company’s API. Currently, these services support two models: Mistral Small and Mistral 7B. The company plans to extend this support to more models in the coming weeks.
Custom Training Services for Select Customers
Mistral is also offering bespoke training services to a select group of customers. This service allows organizations to fine-tune any Mistral model using their own data, creating highly specialized and optimized models for specific applications. “This approach enables the creation of highly specialized and optimized models for their specific domain,” the company notes in a blog post.
Mistral, which is reportedly seeking to raise around $600 million at a $6 billion valuation with backing from investors like DST, General Catalyst, and Lightspeed Venture Partners, is aiming to boost its revenue streams as competition in the generative AI market intensifies.
Since launching its first generative model in September 2023, Mistral has introduced several more models, including one for code generation, and has begun offering paid APIs. However, the company has not yet revealed user numbers or revenue figures.
See also: Raspberry Pi And Hailo Team Up For New AI Extension Kit
