[Bugfix] Mistral Devstral tool calling

[vvidovic]

Mistral's Codestral model mistralai/Devstral-Small-2505 fails when using with tools.
Similar to the existing patch for assistant message content (converting to string) we must convert tool message content to string + make sure the reasoning_content is not present.

FIX Tools usage with mistralai/Devstral-Small-2505 failsMistral

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[mgoin]

Could you please add a test for the new model behavior?

[vvidovic]

Could you please add a test for the new model behavior?

I will just close the issue, it is to hard to run locally the one python test that I would like to update.
After spending many hours updating nvidia drivers, cuda toolkit and similar, I still wasn't able to run successfully pytest tests/mistral_tool_use/test_mistral_tool_calls.py in my local venv environment (Ubuntu 22.04). Maybe it is the result of the Python version (3.10.12).
I don't have more time for this "15 minutes fix".
Hopefully, someone who has already set up a dev environment will solve this issue and I will use the patched wheel of vllm 0.8.5 in the meantime.

[mgoin]

Okay thanks for letting me know, I can take it on

[mgoin]

Superseded by #18729

[vvidovic]

Okay thanks for letting me know, I can take it on

Thanks a lot.

I did try again yesterday by reinstalling nvidia drivers and cuda toolkit and using uv venv instead of pip -m venv to create new environment. It worked on my laptop but my 4 GiB GPU memory was not enough to run the test.
I also tried to run test in the VM in the cloud with the L40S GPU but I was stuck again on the test message "Failed to infer device type".