[Bug]: `guided_regex` not working on M2 Ultra VLLM

[NilsHellwig]

Your current environment

Python 3.12, used conda and official installation guide.

Happens regardless of the employed model.

🐛 Describe the bug

The output of `python -m vllm.entrypoints.openai.api_server --trust-remote-code --model google/gemma-3-4b-it --port 27090 --max-model-len 10000 --api-key token-abc123`

In Console:

WARNING 06-16 09:21:02 [__init__.py:34] xgrammar module cannot be imported successfully. Falling back to use outlines instead.

Happens regardless of the employed model.

Before submitting a new issue...

• [x] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[aarnphm]

fyi https://github.com/vllm-project/vllm/pull/18359

but structured outputs in CPU has limited suppport.

[NilsHellwig]

But I'm using the Mac's GPU

[aarnphm]

vllm doesn't have support with mlx, so not sure it is even being used there.