smaller dockerfile

Author: spa5k

Dockerfile

@@ -1,98 +1,56 @@
-FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim AS builder
-ARG CPU_ONLY=false
+FROM python:3.12-slim-bookworm
 
-WORKDIR /app
+# Install system dependencies
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    libgl1 libglib2.0-0 curl wget git procps \
+    && rm -rf /var/lib/apt/lists/*
 
-# Install build dependencies
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends libgl1 libglib2.0-0 && \
-    rm -rf /var/lib/apt/lists/*
+# Copy UV from official image
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
 
-# Enable bytecode compilation and set proper link mode for cache mounting
 ENV UV_COMPILE_BYTECODE=1 \
     UV_LINK_MODE=copy \
-    HF_HOME=/app/.cache/huggingface \
-    TORCH_HOME=/app/.cache/torch \
-    PYTHONPATH=/app \
+    UV_SYSTEM_PYTHON=1 \
+    HF_HOME=/tmp/ \
+    TORCH_HOME=/tmp/ \
     OMP_NUM_THREADS=4
 
-# Copy dependency files and README
-COPY pyproject.toml uv.lock README.md ./
+WORKDIR /app
 
-# Install dependencies but not the project itself
+RUN echo "# Docling API" > README.md
+
+# Install dependencies first (for better layer caching)
+COPY pyproject.toml uv.lock ./
 RUN --mount=type=cache,target=/root/.cache/uv \
     uv sync --frozen --no-install-project
 
-# Copy the rest of the project
-COPY . .
-
-# Better GPU detection: Check both architecture and if NVIDIA is available
-RUN ARCH=$(uname -m) && \
-    if [ "$CPU_ONLY" = "true" ] || [ "$ARCH" = "aarch64" ] || [ "$ARCH" = "arm64" ] || ! command -v nvidia-smi >/dev/null 2>&1; then \
-    USE_GPU=false; \
-    else \
-    USE_GPU=true; \
-    fi && \
-    echo "Detected GPU availability: $USE_GPU" && \
-    # For PyTorch installation with architecture detection
-    uv pip uninstall -y torch torchvision torchaudio || true && \
-    if [ "$USE_GPU" = "false" ]; then \
-    # For CPU or ARM architectures or no NVIDIA
-    echo "Installing PyTorch for CPU" && \
-    uv pip install --no-cache-dir torch torchvision --extra-index-url https://download.pytorch.org/whl/cpu; \
+ARG CPU_ONLY=false
+RUN if [ "$CPU_ONLY" = "true" ]; then \
+    uv pip install --system --no-cache-dir torch torchvision --extra-index-url https://download.pytorch.org/whl/cpu; \
     else \
-    # For x86_64 with GPU support
-    echo "Installing PyTorch with CUDA support" && \
-    uv pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121; \
+    uv pip install --system --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121; \
     fi
 
-# Install the project in non-editable mode
+# Install required packages
 RUN --mount=type=cache,target=/root/.cache/uv \
-    uv sync --frozen --no-editable
-
-# Download models for the pipeline
-RUN uv run python -c "from docling.pipeline.standard_pdf_pipeline import StandardPdfPipeline; artifacts_path = StandardPdfPipeline.download_models_hf(force=True)"
+    uv pip install --system --no-cache-dir docling easyocr
 
-# Pre-download EasyOCR models with better GPU detection
-RUN ARCH=$(uname -m) && \
-    if [ "$CPU_ONLY" = "true" ] || [ "$ARCH" = "aarch64" ] || [ "$ARCH" = "arm64" ] || ! command -v nvidia-smi >/dev/null 2>&1; then \
-    echo "Downloading EasyOCR models for CPU" && \
-    uv run python -c "import easyocr; reader = easyocr.Reader(['fr', 'de', 'es', 'en', 'it', 'pt'], gpu=False); print('EasyOCR CPU models downloaded successfully')"; \
-    else \
-    echo "Downloading EasyOCR models with GPU support" && \
-    uv run python -c "import easyocr; reader = easyocr.Reader(['fr', 'de', 'es', 'en', 'it', 'pt'], gpu=True); print('EasyOCR GPU models downloaded successfully')"; \
-    fi
-
-# Production stage
-FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim
-WORKDIR /app
-
-# Install runtime dependencies
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends redis-server libgl1 libglib2.0-0 curl && \
-    rm -rf /var/lib/apt/lists/*
+# Download models in a single step
+RUN python -c 'from docling.pipeline.standard_pdf_pipeline import StandardPdfPipeline; \
+    from easyocr import Reader; \
+    artifacts_path = StandardPdfPipeline.download_models_hf(force=True); \
+    reader = Reader(["fr", "de", "es", "en", "it", "pt"], gpu=True); \
+    print("Models downloaded successfully")'
 
-# Set environment variables
-ENV HF_HOME=/app/.cache/huggingface \
-    TORCH_HOME=/app/.cache/torch \
-    PYTHONPATH=/app \
-    OMP_NUM_THREADS=4 \
-    UV_COMPILE_BYTECODE=1
-
-# Create a non-root user
-RUN useradd --create-home app && \
-    mkdir -p /app && \
-    chown -R app:app /app /tmp
-
-# Copy the virtual environment from the builder stage
-COPY --from=builder --chown=app:app /app/.venv /app/.venv
-ENV PATH="/app/.venv/bin:$PATH"
+# Copy the application code
+COPY . .
 
-# Copy necessary files for the application
-COPY --chown=app:app . .
+# Final dependency sync
+RUN --mount=type=cache,target=/root/.cache/uv uv sync --frozen
 
-# Switch to non-root user
-USER app
+# Remove cache to save space
+RUN rm -rf /root/.cache/uv
 
 EXPOSE 8080
-CMD ["uvicorn", "main:app", "--port", "8080", "--host", "0.0.0.0"]
+
+CMD ["uv", "run", "uvicorn", "--port", "8080", "--host", "0.0.0.0", "main:app"]