Mistral Unveils Enhanced AI Model Customization and SDK

French AI startup Mistral has launched new services and a software development kit (SDK).

These offerings enable developers and enterprises to fine-tune its generative models for specific use cases. They also include both self-service and managed solutions, catering to a wide range of customization needs.

Self-Service Fine-Tuning with Mistral-Finetune

Mistral has introduced Mistral-Finetune, an SDK designed for fine-tuning its AI models on workstations, servers, and small datacenter nodes. This SDK, available on GitHub, is optimized for multi-GPU setups but can also function with a single Nvidia A100 or H100 GPU for smaller models like Mistral 7B. According to Mistral, fine-tuning a dataset such as UltraChat, which contains 1.4 million dialogs with OpenAI’s ChatGPT, takes approximately 30 minutes using Mistral-Finetune across eight H100 GPUs.

Managed Fine-Tuning Services

For those preferring a managed solution, Mistral now offers fine-tuning services accessible via the company’s API. These services currently support Mistral Small and Mistral 7B models, with plans to extend support to more models in the coming weeks. This option is ideal for developers and companies seeking a more streamlined and hassle-free customization process.

Custom Training Services for Enterprises

Mistral is also rolling out custom training services for select customers, allowing fine-tuning of any Mistral model using an organization’s specific data. This service aims to create highly specialized and optimized models tailored to particular domains, enhancing the applicability and performance of Mistral’s AI solutions in various industries.

Strategic Growth and Future Plans

Mistral, which is reportedly seeking to raise approximately $600 million at a $6 billion valuation from investors including DST, General Catalyst, and Lightspeed Venture Partners, is aiming to expand its revenue streams amidst rising competition in the generative AI sector. Since launching its first generative model in September 2023, Mistral has introduced several models, including one for code generation, and has implemented paid APIs. However, the company has yet to disclose user numbers or revenue figures.

These new services and the SDK reflect Mistral’s commitment to providing versatile and robust AI solutions, positioning itself as a formidable player in the evolving AI landscape.