Update Dockerfile

Dockerfile

@@ -1,21 +1,31 @@
-# Use the official Hugging Face GGUF server image
-# This comes pre-compiled and optimized for CPU
-FROM ghcr.io/huggingface/llama-cpp-gguf-server:latest
+# Use Python 3.10 to match the pre-compiled wheel
+FROM python:3.10-slim
 
-# Set environment variables for the server
-ENV MODEL_ID="HauhauCS/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive"
-ENV MODEL_FILE="Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"
-ENV PORT=7860
-ENV HOST=0.0.0.0
+WORKDIR /app
 
-# The server automatically handles the download and OpenAI endpoint mapping
+# Install system essentials
+RUN apt-get update && apt-get install -y \
+    gcc g++ make cmake git libopenblas-dev wget \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install llama-cpp-python using a pre-built HF Space wheel (Skips 30min build)
+# This wheel is optimized for the 2-vCPU free tier
+RUN pip install --no-cache-dir \
+    https://huggingface.co/Luigi/llama-cpp-python-wheels-hf-spaces-free-cpu/resolve/main/llama_cpp_python-0.3.22-cp310-cp310-linux_x86_64.whl
+
+# Install the server and hub tools
+RUN pip install --no-cache-dir "llama-cpp-python[server]" huggingface_hub
+
+# Download the model during build so startup is instant
+RUN python3 -c "from huggingface_hub import hf_hub_download; hf_hub_download(repo_id='HauhauCS/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive', filename='Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf', local_dir='.')"
+
+# Expose the HF Space port
 EXPOSE 7860
 
-# Run the server with CPU-optimized threads
-CMD ["llama-server", \
-     "--hf-repo", "HauhauCS/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive", \
-     "--hf-file", "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf", \
+# Run the OpenAI-compatible server
+CMD ["python3", "-m", "llama_cpp.server", \
+     "--model", "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf", \
      "--host", "0.0.0.0", \
      "--port", "7860", \
-     "--threads", "2", \
-     "--ctx-size", "2048"]
+     "--n_threads", "2", \
+     "--n_ctx", "2048"]