22 lines
631 B
Docker
22 lines
631 B
Docker
|
FROM harbor.4pd.io/dooke/vllm/vllm/vllm-openai:v0.11.0
|
||
|
|
|
||
|
|
# 修复1 transformers 库级 patch
|
||
|
|
COPY patch.py /tmp/patch.py
|
||
|
|
RUN python3 /tmp/patch.py
|
||
|
|
|
||
|
|
# 修复2 Triton backend head_size 验证绕过
|
||
|
|
COPY patch_triton.py /tmp/patch_triton.py
|
||
|
|
RUN python3 /tmp/patch_triton.py
|
||
|
|
|
||
|
|
# 修复3 运行时 tokenizer 配置修复脚本
|
||
|
|
COPY detect_tokenizer.py /opt/detect_tokenizer.py
|
||
|
|
COPY fix_tokenizer.py /opt/fix_tokenizer.py
|
||
|
|
|
||
|
|
# 修复4 head_size 检测，自动切换 attention backend
|
||
|
|
COPY detect_head_size.py /opt/detect_head_size.py
|
||
|
|
|
||
|
|
COPY entrypoint.sh /opt/entrypoint.sh
|
||
|
|
RUN chmod +x /opt/entrypoint.sh
|
||
|
|
|
||
|
|
ENTRYPOINT ["/opt/entrypoint.sh"]
|