diff --git a/Dockerfile b/Dockerfile index 7784c2a..ec5b10e 100644 --- a/Dockerfile +++ b/Dockerfile @@ -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"] \ No newline at end of file diff --git a/Dockerfile.qwen3-14b b/Dockerfile.qwen3-32b similarity index 98% rename from Dockerfile.qwen3-14b rename to Dockerfile.qwen3-32b index c1861ad..7784c2a 100644 --- a/Dockerfile.qwen3-14b +++ b/Dockerfile.qwen3-32b @@ -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"] \ No newline at end of file