fix bug

Dockerfile

@@ -31,7 +31,13 @@ COPY --chown=user ./requirements.txt requirements.txt
 # --prefer-binary : never try to build any wheel from source (flash_attn
 #                   would OOM the build container). The base image already
 #                   provides torch / vllm / flash_attn / transformers.
-RUN pip install --no-cache-dir --upgrade --prefer-binary -r requirements.txt
+#
+# NOTE: deliberately NO `--upgrade` here. The base image ships a carefully
+# pinned dependency tree (transformers 4.x compatible with
+# huggingface-hub<1.0). Using --upgrade lets pip pull huggingface-hub 1.x
+# whenever any package's metadata allows it, which then breaks
+# `import vllm` at container start.
+RUN pip install --no-cache-dir --prefer-binary -r requirements.txt
 
 COPY --chown=user . /app
 CMD ["python3", "app.py"]

requirements.txt

@@ -7,6 +7,14 @@
 #   Re-installing any of them here triggers a full source build of flash_attn
 #   (no PyPI wheel available), which OOM-kills the HF Space *build*
 #   container. Keep this list to lightweight, app-only dependencies.
+#
+# IMPORTANT version pin:
+#   The transformers shipped in vllm 0.8.5 requires huggingface-hub<1.0.
+#   Without an explicit upper bound here, pip happily upgrades
+#   huggingface-hub to 1.x (released 2025-Q4), which then makes
+#   `import transformers` -> `import vllm` fail at container start with:
+#     ImportError: huggingface-hub>=0.30.0,<1.0 is required ...
+huggingface-hub>=0.30.0,<1.0
 librosa>=0.11.0
 nnAudio>=0.3.3
 gradio>=4.44