Mistral Introduces New Customization Options for AI Model Fine-Tuning
Introduction
French AI startup Mistral is offering new customization options for developers and enterprises to fine-tune its generative models for specific use cases. These options include self-service SDK, managed services through API, and custom training services.
Self-Service SDK: Mistral-Finetune
- Mistral has released Mistral-Finetune, a software development kit (SDK) for fine-tuning its models.
- The SDK is optimized for multi-GPU setups but can scale down to a single Nvidia A100 or H100 GPU for smaller models.
- Fine-tuning on datasets like UltraChat takes around half an hour using Mistral-Finetune across multiple GPUs.
Managed Services through API
- For developers and companies preferring a managed solution, Mistral offers fine-tuning services through its API.
- Initially compatible with Mistral Small and Mistral 7B models, more model support is expected in the future.
Custom Training Services
- Mistral now provides custom training services for select customers to fine-tune any Mistral model using their data.
- This approach enables the creation of specialized and optimized models for specific domains.
Growth Strategy
- Mistral is seeking a funding round of $600 million at a $6 billion valuation from investors including DST, General Catalyst, and Lightspeed Venture Partners.
- To increase revenue, Mistral is expanding its offerings in the generative AI space, with new models and paid APIs.
Conclusion
- Mistral continues to innovate in the AI space with new customization options for fine-tuning its generative models.
- With a focus on self-service SDK, managed services through API, and custom training services, Mistral aims to cater to the evolving needs of developers and enterprises in the AI landscape.
