sirius/gcloud-lab
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix(vllm): update to recommended qwen3_coder tool parser, auto max-model-len, and latest image

Browse source 
Addresses tokenizer backend / parser mismatch on heretic-style Qwen3.5 models per model card recommendations.
This commit is contained in:
Sirius Claw 2026-04-22 06:32:35 +00:00
parent e7debc708c
commit 3133a3b32f
1 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
apps/base/customer1/openclaw/a100-vllm.yaml
View file

@ -24,7 +24,7 @@ spec:
effect: "NoSchedule"
containers:
- name: vllm-brain
image: vllm/vllm-openai:v0.19.1-ubuntu2404
image: vllm/vllm-openai:latest
command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
env:
- name: HF_TOKEN
@ -36,12 +36,12 @@ spec:
- --tensor-parallel-size=1
- --gpu-memory-utilization=0.95
- --kv-cache-dtype=fp8
- --max-model-len=49152
- --max-model-len=auto
- --enable-auto-tool-choice
- --trust-remote-code
- --dtype=auto
- --enable-prefix-caching
- --tool-call-parser=hermes
- --tool-call-parser=qwen3_coder
- --reasoning-parser=qwen3
- --disable-custom-all-reduce