French AI company Mistral is launching new customization features for its AI models, including paid plans, allowing developers and businesses to tailor the generative models to specific requirements.
The first option is self-service. Mistral has introduced a software development kit (SDK), Mistral-Finetune, for fine-tuning its models on various platforms, including workstations, servers, and small datacenter nodes.
According to the readme file in the SDK’s GitHub repository, the SDK is optimized for multi-GPU configurations but can also be used with a single Nvidia A100 or H100 GPU for tuning smaller models such as Mistral 7B. Fine-tuning on a dataset like UltraChat, which consists of 1.4 million dialogues with OpenAI’s ChatGPT, takes about 30 minutes using Mistral-Finetune across eight H100s, as per Mistral.
For those who prefer a managed solution, Mistral is also offering new fine-tuning services through its API. Currently compatible with two models, Mistral Small and Mistral 7B, Mistral plans to extend this support to more models in the coming weeks.
Additionally, Mistral is introducing custom training services, which are currently available only to select customers. These services allow any Mistral model to be fine-tuned for specific applications using an organization’s data. “This approach enables the creation of highly specialized and optimized models for their specific domain,” the company states in a post on its official blog.
Mistral, which my colleague Ingrid Lunden recently reported is seeking to raise around $600 million at a $6 billion valuation from investors such as DST, General Catalyst, and Lightspeed Venture Partners, is aiming to increase its revenue amid significant — and growing — competition in the generative AI market.
Since unveiling its first generative model in September 2023, Mistral has released several more models, including a code-generating model, and introduced paid APIs. However, it has not revealed the number of its users or its revenue figures.
