with venv to resolve errors

Dockerfile

@@ -3,27 +3,28 @@ FROM ghcr.io/abetlen/llama-cpp-python:latest
 
 WORKDIR /app
 
-# The base image already has llama-cpp-python,
-# so we only need to install huggingface_hub
-RUN pip install huggingface_hub
-
-# --- Model Download ---
 # Copy the download script into the container
 COPY download_model.py .
 
 # Make the HF_TOKEN secret available as an argument
-# This will be passed in by the HF Spaces platform
 ARG HF_TOKEN
-# Run the script to download the model
+
+# --- FIX: Use a temporary virtual env to install dependencies ---
+# This creates a venv, installs huggingface_hub inside it,
+# runs the download script, and then this venv is discarded.
+# This prevents our pip install from breaking the base image.
 RUN --mount=type=secret,id=HF_TOKEN \
+    python -m venv /tmp/downloader-venv && \
+    . /tmp/downloader-venv/bin/activate && \
+    pip install huggingface_hub && \
     python download_model.py
 
 # --- Server Runtime ---
 # Expose port 8000 (which we defined in README.md)
 EXPOSE 8000
 
-# This is the command that will run when the container starts
-# It reads the API_KEY secret from the environment
+# This command runs in the base image's original environment
+# which should be stable and correctly linked.
 CMD [ \
     "python", \
     "-m", "llama_cpp.server", \