fix(mistral): guard ReasoningEffort import for older mistral_common versions

[hijingsong]

Fix: Guard ReasoningEffort import for older mistral_common versions

Fixes #45372

Problem

ReasoningEffort was added in mistral-common>=1.10.0, but the import in tokenization_mistral_common.py was unconditional within the is_mistral_common_available() guard. Users with mistral-common<1.10.0 would get an ImportError that prevented loading any processor — including non-Mistral models like Gemma 4 and Qwen.

Additionally, even stubbing ReasoningEffort = None was insufficient because line 1042 uses ReasoningEffort | None in a type annotation, causing TypeError: unsupported operand type(s) for | since NoneType does not support the | operator.

Fix

Wrap the ReasoningEffort import in a try/except ImportError, falling back to typing.Any. This ensures:

1. The module loads regardless of mistral-common version
2. The ReasoningEffort | None type annotation resolves correctly (Any | None is valid)
3. Non-Mistral models work without requiring the latest mistral-common
4. Mistral models with reasoning_effort support work correctly when mistral-common>=1.10.0 is installed

Changes

• src/transformers/tokenization_mistral_common.py: Split the combined import of ChatCompletionRequest and ReasoningEffort into separate imports with a try/except guard

Testing

• Syntax validation passes
• Type annotation Any | None resolves correctly at runtime
• No changes to runtime behavior for users with mistral-common>=1.10.0

[zucchini-nlp]

We have pinned it to 1.10.0 so ig you need to re-install to get the updated version

transformers/setup.py

Line 96 in def8e6a

"mistral-common[image]>=1.10.0",