Mistral AI has unveiled its new model customization on La Plateforme. Called mistral-finetune, the update will provide fine-tuning capabilities to enhance performance, speed, and editorial control of its users’ artificial intelligence (AI) applications.
According to an announcement on Mistral AI’s website, the recently released mistral-finetune allows its users to adapt its open-source AI models to fine-tune their infrastructure. The France-based company describes it as a lightweight and memory-efficient codebase built on the Low-Rank Adaptation (LoRA) training paradigm.
In addition, Mistral AI will offer services on La Plateforme to provide developers with infrastructure fine-tuning support. The services are based upon the company’s fine-tuning techniques, which, the company says, makes model adaptation and deployment more cost-efficient and effective.
“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,” Mistral AI said in the announcement.
The model fine-tuning services are compatible with Mistral 7B and Mistral Small, allowing users leveraging one of these models to use Mistral’s application programming interface (API) to tailor the model. It also plans to introduce additional models to its fine-tuning services in the near future.
According to a previous announcement, Mistral AI opened up beta access to its La Plateforme in December 2023, with three chat endpoints for text generation following textual instructions and an embedding endpoint.
