8 Commits
Author SHA1 Message Date
wanglin 0ff012882c revert: 撤销 10-17 误推的 FROM 改动,恢复 a3a160df(10-10) 的 harbor 基础镜像
10-17 那次推送依据「官方仓库用 git.modelhub.org.cn:9443」把 FROM 改回内网地址,
但按 blob 实测:16 次 A 层「Docker 镜像构建失败」发生在 10-01~10-04,当时 FROM 正是该内网地址;
a3a160df(10-10 01:35) 才把它换成 harbor.4pd.io。故那次推送是撤销修复、属回归,现予回退。
边界:harbor 版从未被任何提交实际跑过,两个地址孰对仍未验证。
2026-10-24 23:56:12 +08:00
wanglin 6e68be0b30 fix: �޸� Dockerfile FROM �У��Ļعٷ�������ַ
A ��ʧ�ܣ�Docker ���񹹽�ʧ�ܣ��ĸ���ԭ����FROM �����˴����ľ�����ַ harbor.4pd.io��
�ٷ�Ӧ���� git.modelhub.org.cn:9443��
2026-10-17 12:14:29 +08:00
wanglin a3a160df82 按官方工单指示替换基础镜像:git.modelhub.org.cn:9443 → harbor.4pd.io/modelhubxc(同 tag v1.2.3,解 Docker 镜像构建失败) 2026-10-10 01:35:57 +08:00
wanglin 8d3ee4a9e3 v3b: A/B对照 — 移除 --hf-overrides 以逐字节对齐成功配置(ouys/sub400)
v3b: A/B对照 — 移除 --hf-overrides 以逐字节对齐成功配置(ouys/sub400)
2026-10-02 20:45:45 +08:00
wanglin ee1c3c4c9b add --hf-overrides architectures=Qwen3_5MoeForCausalLM (fix model config arch mismatch) 2026-10-02 20:00:19 +08:00
wanglin 941c42a04f v3: 对齐成功配置 — 补 --enforce-eager 与 corex 运行库环境变量, max-model-len 256K / gmu 0.95 / seqs 4 / bt 8192
v3: 对齐成功配置 — 补 --enforce-eager 与 corex 运行库环境变量, max-model-len 256K / gmu 0.95 / seqs 4 / bt 8192
2026-10-02 19:53:25 +08:00
wanglin d6dfff0918 v2: conservative max-num-seqs=8 + batched-tokens=16384, revert dtype/256K (v1 OOM crash) 2026-10-01 23:29:16 +08:00
wanglin a56835199e 更新 computility-run.yaml
优化启动参数:max-model-len→256K, max-num-seqs→8, batched-tokens→16384, 加prefix-caching和dtype
2026-09-24 07:57:33 +08:00
2 changed files with 54 additions and 33 deletions

No files matched your search

+2 -2
View File
@@ -1,6 +1,6 @@
FROM git.modelhub.org.cn:9443/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3
FROM harbor.4pd.io/modelhubxc/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3
RUN mkdir /workspace
WORKDIR /workspace/
COPY ./qwen3_6_scripts /workspace/qwen3_6_scripts
RUN cd ./qwen3_6_scripts && ./patch_ops.sh
RUN cd ./qwen3_6_scripts && ./patch_ops.sh
+52 -31
View File
@@ -1,34 +1,55 @@
concurrency: 1
command:
- python3
- -m
- vllm.entrypoints.openai.api_server
- --model
- /model
- --served-model-name
- llm
- --max-model-len
- '100000'
- --gpu-memory-utilization
- '0.9'
- --trust-remote-code
- -tp
- '4'
- --max-num-seqs
- '1'
- --disable-log-requests
- --disable-frontend-multiprocessing
- --max-num-batched-tokens
- '8192'
- --enable-chunked-prefill
- --max-seq-len-to-capture
- '32768'
- --enable-auto-tool-choice
- --tool-call-parser
- qwen3_coder
- --reasoning-parser
- qwen3
- --enable-prefix-caching
- python3
- -m
- vllm.entrypoints.openai.api_server
- --model
- /model
- --served-model-name
- llm
- --max-model-len
- '256000'
- --gpu-memory-utilization
- '0.95'
- --trust-remote-code
- -tp
- '4'
- --max-num-seqs
- '4'
- --disable-log-requests
- --disable-frontend-multiprocessing
- --max-num-batched-tokens
- '8192'
- --enable-chunked-prefill
- --max-seq-len-to-capture
- '32768'
- --enforce-eager
- --enable-auto-tool-choice
- --tool-call-parser
- qwen3_coder
- --reasoning-parser
- qwen3
- --enable-prefix-caching
- --dtype
- half
env:
- name: VLLM_ENGINE_ITERATION_TIMEOUT_S
value: 3600
- 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
- name: VLLM_COREX_FA2_LIBRARY
value: /usr/local/corex/lib64/libcorex_fa2.so
- name: VLLM_COREX_GDN_LIBRARY
value: /usr/local/corex/lib64/libcorex_gdn.so
- name: VLLM_COREX_MOE_LIBRARY
value: /usr/local/corex/lib64/libcorex_moe.so
- name: VLLM_REQUEST_METRICS_FILE
value: /tmp/vllm-request-metrics.jsonl
- name: VLLM_CACHE_BLOCK_SIZE
value: 16