Mistral Unveils Enhanced Services and SDK for Customers to Fine-Tune AI Models

French AI startup Mistral is unveiling new customization options for its AI models, including paid plans, aimed at helping developers and enterprises tailor its generative models to specific applications.

The first offering is a self-service solution. Mistral has launched the Mistral-Finetune software development kit (SDK), enabling users to fine-tune models on workstations, servers, and small data center nodes.

In the readme on the SDK’s GitHub repository, Mistral highlights that the SDK is designed for multi-GPU configurations but can efficiently scale down to work with a single Nvidia A100 or H100 GPU when fine-tuning smaller models like Mistral 7B. For instance, fine-tuning using a dataset like UltraChat—which consists of 1.4 million dialogues generated from OpenAI’s ChatGPT—can be completed in approximately 30 minutes using Mistral-Finetune with eight H100 GPUs, according to Mistral.

For developers and businesses seeking a more hands-off approach, Mistral now offers fine-tuning services through the company’s API. Currently compatible with two models—Mistral Small and Mistral 7B—these fine-tuning services will expand to support additional models in the upcoming weeks.

Additionally, Mistral is launching custom training services, available to select clients, allowing organizations to fine-tune any Mistral model using their proprietary data. “This approach creates highly specialized and optimized models tailored to specific domains,” the company states in a recent blog post.

As reported by my colleague Ingrid Lunden, Mistral is aiming to raise around $600 million at a $6 billion valuation, with backing from investors like DST, General Catalyst, and Lightspeed Venture Partners. Mistral is poised for revenue growth as it faces increasing competition in the generative AI landscape.

Since its debut generative model in September 2023, Mistral has released several additional models, including one for code generation, and introduced paid APIs. However, the company has yet to disclose its user count or revenue figures.