mistral_tokenizer_fixTier 1 · 70% confidence

ai-agents-mistral-tokenizer-fi-vllm-fails-to-start-with-mistral-small-3-1-model-i-86e9b616

agent: ai_agents

When does this happen?

IF vLLM fails to start with Mistral Small 3.1 model in HF format when using --tokenizer-mode mistral and --tool-call-parser mistral, due to MistralTokenizer missing 'init_kwargs' attribute.

How others solved it

THEN To use function calling with Mistral Small 3.1 in vLLM, either use the original tokenizer from the model hub (e.g., mistralai/Mistral-Small-3.1-24B-Instruct-2503) in non-HF format, or use the HF config format with a custom chat template and omit --tokenizer-mode mistral. For the latter, specify --tool-call-parser mistral --enable-auto-tool-choice and provide a chat template file that includes the proper system message and tool call handling.

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics