Dockerfile

# ═══════════════════════════════════════════════════════════════════
#  Minecraft Bot LLM Backend — Dockerfile
#  Target: Hugging Face Spaces (Docker template), port 7860
#  Model:  Qwen2.5-Coder-7B-Instruct (Q4_K_M GGUF, ~4.4 GB)
#  Runtime: llama-cpp-python (CPU-optimised build)
# ═══════════════════════════════════════════════════════════════════

# ── Stage 1: builder ────────────────────────────────────────────────
FROM python:3.11-slim AS builder

# Build deps for llama-cpp-python C++ compilation
RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    cmake \
    git \
    wget \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /build

COPY requirements.txt .
# Build llama-cpp-python wheel with optimised CPU flags (AVX2 if available)
# Set CMAKE_ARGS for pure CPU; swap for CUDA_DOCKER_ARCH if GPU available
ENV CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS"
ENV FORCE_CMAKE=1

RUN pip install --upgrade pip && \
    pip install --no-cache-dir -r requirements.txt --target /build/deps

# ── Stage 2: runtime ────────────────────────────────────────────────
FROM python:3.11-slim

# Runtime system libs
RUN apt-get update && apt-get install -y --no-install-recommends \
    libgomp1 \
    libopenblas-dev \
    wget \
    && rm -rf /var/lib/apt/lists/*

# ── HF Spaces requires a non-root user named "user" ─────────────────
RUN useradd -m -u 1000 user
USER user

ENV HOME=/home/user \
    PATH=/home/user/.local/bin:$PATH \
    PYTHONPATH=/app/deps

WORKDIR /app

# Copy installed Python packages from builder
COPY --from=builder --chown=user /build/deps /app/deps

# Copy application source
COPY --chown=user app.py .

# ── Prepare model directory ──────────────────────────────────────────
# Tạo sẵn thư mục lưu model với quyền của user hiện tại
RUN mkdir -p /app/models

# ── Environment defaults (overridable via HF Space Secrets) ─────────
ENV MODEL_PATH=/app/models/qwen2.5-coder-7b-instruct-q4_k_m.gguf \
    MODEL_NAME=qwen2.5-coder-7b-instruct \
    N_CTX=4096 \
    N_THREADS=4 \
    N_GPU_LAYERS=0 \
    MAX_TOKENS=1024
# BEARER_TOKEN must be injected as a Secret — never hard-coded here

EXPOSE 7860

CMD ["python", "-m", "uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]