The company also offers an embedding endpoint, mistral-embed, which serves an embedding model designed for retrieval capabilities. The APIs follow the specifications of a popular chat interface and provide Python and Javascript client libraries for querying. As the platform moves from beta access to general availability, anyone can register to use the API. The platform is supported by Nvidia, which has assisted in integrating TRT-LLM and Triton.
Key takeaways:
- Mistral AI is launching beta access to its platform services, which include three chat endpoints for generating text and an embedding endpoint.
- The company offers three generative endpoints: Mistral-tiny, Mistral-small, and Mistral-medium, each with different performance and price tradeoffs. These models are trained on data from the open web and fine-tuned for instructions.
- Mistral-embed, the embedding endpoint, serves an embedding model with a 1024 embedding dimension, designed with retrieval capabilities in mind.
- Anyone can register to use Mistral AI's API as the company ramps up its capacity, with the business team available to help qualify needs and accelerate access.