how to disable muse glimmer thinking? on vllm
There is currently no documented way to disable “thinking” (reasoning) mode for Muse Glimmer in vLLM. Unlike models such as Qwen3, which support disabling thinking via chat template kwargs (e.g., “enable_thinking”: false), Muse Glimmer does not expose such a toggle in the vLLM configuration or API, and this limitation is confirmed in user reports and issue discussions. Attempts to use similar parameters or omit the reasoning parser do not disable thinking for Muse Glimmer, and this is considered a model or implementation constraint, not a vLLM bug or missing feature (Bug: Using vLLM to deploy Minimax m2.5, the thinking/reasoning cannot be disable; Add Muse Glimmer model support).
Would you like more detail on how this compares to other models or on the technical reasons for this limitation?
Sources:
- Bug: Using vLLM to deploy Minimax m2.5, the thinking/reasoning cannot be disable
- Add Muse Glimmer model support
Hint: Mention @Herald in the post for followups.