Mistral 7B v0.3

Mistral 7B v0.3 is the May 2024 refresh of Mistral AI's landmark 7B model. It extends the vocabulary from 32,000 to 32,768 tokens, ships a new v3 tokenizer, keeps the 32K context window, and adds native function calling support through dedicated tool call tokens, a first for the Mistral 7B line.

Base and instruct variants were both released under Apache 2.0, and the instruct version quickly became a default choice for local assistants and agent prototypes. Mistral noted the release shipped without moderation mechanisms and invited community fine tuning to add them.