hailin
/
sglang.0.4.8.post1
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

.

This commit is contained in:
hailin 2025-07-07 15:24:18 +08:00
parent c5e4ef4a6d
commit a2a93c7c4c
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Dockerfile
View File

@ -160,7 +160,7 @@ RUN chmod +x /tini
ENTRYPOINT ["/tini", "--"]
# ---- 拷贝模型（路径可换） ----
COPY ./Alibaba/Qwen3-32B /root/.cradle/Alibaba/Qwen3-32B
COPY ./Alibaba/Qwen3-14B /root/.cradle/Alibaba/Qwen3-14B
HEALTHCHECK --interval=30s --timeout=2s --start-period=300s --retries=5 CMD curl -fs http://localhost:30000/health || exit 1
@ -171,7 +171,7 @@ EXPOSE 30000
CMD ["python3", "-m", "sglang.launch_server", \
"--host", "0.0.0.0", \
"--port", "30000", \
"--model-path", "/root/.cradle/Alibaba/Qwen3-32B/", \
"--tp", "4", \
"--model-path", "/root/.cradle/Alibaba/Qwen3-14B/", \
"--tp", "2", \
"--api-key", "token-abc123", \
"--enable-metrics"]

6
Dockerfile.qwen3-14b → Dockerfile.qwen3-32b
View File

@ -160,7 +160,7 @@ RUN chmod +x /tini
ENTRYPOINT ["/tini", "--"]
# ---- 拷贝模型（路径可换） ----
COPY ./Alibaba/Qwen3-14B /root/.cradle/Alibaba/Qwen3-14B
COPY ./Alibaba/Qwen3-32B /root/.cradle/Alibaba/Qwen3-32B
HEALTHCHECK --interval=30s --timeout=2s --start-period=300s --retries=5 CMD curl -fs http://localhost:30000/health || exit 1
@ -171,7 +171,7 @@ EXPOSE 30000
CMD ["python3", "-m", "sglang.launch_server", \
"--host", "0.0.0.0", \
"--port", "30000", \
"--model-path", "/root/.cradle/Alibaba/Qwen3-14B/", \
"--tp", "1", \
"--model-path", "/root/.cradle/Alibaba/Qwen3-32B/", \
"--tp", "4", \
"--api-key", "token-abc123", \
"--enable-metrics"]