Concept
enable_thinking
A switch variable in the chat templates of some models (such as the Qwen series) that controls whether the output includes a thinking section.
Timeline (last 90 days)
Oct 19
On the vLLM endpoint, disabling thinking via chat_template_kwargs: { enable_thinking: false } took 65 seconds, whereas sending reasoning_effort none took 2 seconds.
Unverified50%