update

Dockerfile

@@ -1,30 +1,28 @@
-# Use a pre-built image that already has llama-cpp-python compiled
-FROM ghcr.io/abetlen/llama-cpp-python:latest
+# Use a standard Python 3.12 image
+FROM python:3.12-slim
 
 WORKDIR /app
 
-# Copy the download script into the container
-COPY download_model.py .
+# --- 1. Install build-essential and cmake ---
+# This is necessary for compiling the C++ code
+RUN apt-get update && apt-get install -y build-essential cmake
 
-# Make the HF_TOKEN secret available as an argument
-ARG HF_TOKEN
+# --- 2. Install Python Dependencies (with CPU-only build) ---
+# We set CMAKE_ARGS to disable CUDA, which makes the
+# build *much* faster and avoids the job timeout.
+ENV CMAKE_ARGS="-DLLAMA_CUDA=OFF"
+RUN pip install "llama-cpp-python[server]" huggingface_hub
 
-# --- 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.
+# --- 3. Model Download ---
+# This part is correct and remains the same.
+COPY download_model.py .
+ARG HF_TOKEN
 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)
+# --- 4. Server Runtime ---
+# This part is also correct and remains the same.
 EXPOSE 8000
-
-# This command runs in the base image's original environment
-# which should be stable and correctly linked.
 CMD [ \
     "python", \
     "-m", "llama_cpp.server", \