My Tailor is Mistral

Today, we're introducing model customization on la Plateforme, to bring performance, speed and fine editorial control to your AI application. You can now easily and efficiently adapt Mistral AI models to your specific needs, reducing the cost and expertise required for customizing generative AI models at scale. Whether you want to fine-tune Mistral AI models on your own infrastructure or leverage our proprietary fine-tuning techniques with our managed fine-tuning services, we've got you covered.

Fine-tuning is a powerful technique for customizing and improving the performance of LLMs, providing better responses, flexibility and efficiency to specific applications. When tailoring a smaller model to suit specific domains or use cases, it offers a way to match the performance of larger models, reducing deployment costs and improving application speed.

We're proud to announce three different entry points for specialising Mistral AI models.

Our custom training services involve fine-tuning Mistral AI models on a customer's specific applications using their own proprietary data. This approach enables the creation of highly specialized and optimized models for their particular domain. We propose advanced techniques such as continuous pretraining to include proprietary knowledge within the model weights themselves.

For developers who want to fine-tune Mistral's open-source models on their infrastructure, we've released mistral-finetune , a lightweight and efficient codebase for fine-tuning.

Our codebase is built on the LoRA training paradigm, which allows for memory-efficient and performant fine-tuning. With mistral-finetune, you can fine-tune all our open-source models on your infrastructure without sacrificing performance or memory efficiency.

To further facilitate fine-tuning without infrastructure hassle, we're introducing new fine-tuning services on la Plateforme. These services leverage Mistral's unique fine-tuning techniques, refined through extensive research and development, to allow fast and cost-effective model adaptation and effective deployment of fine-tuned models. It's a new step in our mission to expose advanced science methods to AI application developers. We use LoRA adapters under the hood to prevent forgetting base model knowledge and allow for efficient serving.