ba0f67e79e
根因：patch_ops.sh部署到VLLM_ROOT(lib64)，但python3 -m vllm走的是 /usr/local/corex/lib/python3/dist-packages/vllm/（未被覆盖的路径） 导致基础镜像原版api_server.py运行，不识别qwen3_coder/reasoning-parser 修复：launch_server.py在import前遍历sys.path所有vllm安装， 用shutil.copy2强制覆盖api_server/cli_args/serving_chat等 然后from vllm.entrypoints.openai.api_server import *启动
51 lines
1.3 KiB
YAML
51 lines
1.3 KiB
YAML
concurrency: 1
|
|
command:
|
|
- python3
|
|
- /workspace/qwen3_6_scripts/launch_server.py
|
|
- --model
|
|
- /model
|
|
- --served-model-name
|
|
- llm
|
|
- --max-model-len
|
|
- '80000'
|
|
- --gpu-memory-utilization
|
|
- '0.95'
|
|
- --trust-remote-code
|
|
- -tp
|
|
- '4'
|
|
- --max-num-seqs
|
|
- '2'
|
|
- --max-num-batched-tokens
|
|
- '4096'
|
|
- --enable-chunked-prefill
|
|
- --disable-log-requests
|
|
- --disable-frontend-multiprocessing
|
|
- --enforce-eager
|
|
- --enable-auto-tool-choice
|
|
- --tool-call-parser
|
|
- qwen3_coder
|
|
- --enable-prefix-caching
|
|
- --max-seq-len-to-capture
|
|
- '8192'
|
|
- --dtype
|
|
- half
|
|
env:
|
|
- name: VLLM_ENGINE_ITERATION_TIMEOUT_S
|
|
value: '3600'
|
|
- name: VLLM_ATTENTION_BACKEND
|
|
value: XFORMERS
|
|
- name: ENABLE_CUSTOM_IPC
|
|
value: '1'
|
|
- name: PYTHONPATH
|
|
value: /usr/local/corex/lib/python3/dist-packages:/usr/local/corex/lib64/python3/dist-packages
|
|
- name: LD_LIBRARY_PATH
|
|
value: /usr/local/corex/lib64:/usr/local/openmpi/lib:/usr/local/corex/lib64/python3/dist-packages/ixformer
|
|
- name: PYTORCH_CUDA_ALLOC_CONF
|
|
value: max_split_size_mb:512
|
|
- name: OMP_NUM_THREADS
|
|
value: '1'
|
|
- name: BI100_MOE_COREX_DIRECT_ROUTED
|
|
value: '1'
|
|
- name: BI100_GDN_COREX_PACKED_DECODE
|
|
value: '1'
|