worker-vllm icon indicating copy to clipboard operation
worker-vllm copied to clipboard

Feature request: environment variables for reasoning

Open hase808 opened this issue 8 months ago • 3 comments

Hi, I am trying to deploy a reasoning model via Runpod Serverless (qwq from Qwen). However, the environment variables for enable-reasoning and reasoning-parser are missing. The flags were found in the official documentation of vllm.

Is it possible to bring these environment variables in a future release?

hase808 avatar Apr 29 '25 15:04 hase808

related commit https://github.com/runpod-workers/worker-vllm/commit/30dd7c1eb50355a894cd600791e72c5afa3dbefe

nokados avatar May 07 '25 15:05 nokados

Wait, it's not complete. I've just spotted the strange commit, where this feature was commented out

nokados avatar May 12 '25 17:05 nokados

@hase808 I use this custom CMD as temporary fix:

/bin/bash -c 'if ! grep -q "enable_reasoning=True" /src/engine.py; then sed -i "s/# enable_reasoning=/enable_reasoning=True, #/" /src/engine.py && sed -i "s/# reasoning_parser=/reasoning_parser=\"deepseek_r1\", #/" /src/engine.py; fi && python3 /src/handler.py'

bbarwik avatar May 20 '25 17:05 bbarwik