config/Dockerfile.gpu

# =============================================================================
# Stack 2.9 GPU Dockerfile
# Multi-stage build for NVIDIA GPU (CUDA 11.8 + cuDNN 8)
# =============================================================================
# Usage:
#   Build:  docker build -f Dockerfile.gpu -t stack-2.9-gpu .
#   Run:    docker compose -f docker-compose.gpu.yml up
#   Or:     docker run --rm --gpus all -p 8000:8000 \
#             -v $(pwd)/base_model_qwen7b:/model:ro \
#             stack-2.9-gpu
# =============================================================================

# -----------------------------------------------------------------------------
# Stage 1: Builder
# Install Python deps into a wheel, then discard the bulk of the build layer.
# -----------------------------------------------------------------------------
FROM python:3.11-slim AS builder

WORKDIR /build

# Install build dependencies
RUN apt-get update && apt-get install -y --no-install-recommends \
        build-essential \
        curl \
    && rm -rf /var/lib/apt/lists/*

# Install PyTorch with CUDA 11.8 support (CPU fallback pip wheel works too)
# Using PyPI index; for air-gapped envs, swap --index-url for a local mirror.
RUN python -m venv /opt/venv \
    && /opt/venv/bin/pip install --upgrade pip setuptools wheel

# Install ML / inference deps
COPY requirements_api.txt .
RUN /opt/venv/bin/pip install --no-cache-dir -r requirements_api.txt

# Install torch with CUDA support
RUN /opt/venv/bin/pip install --no-cache-dir \
        torch==2.1.2 \
        torchvision==0.16.2 \
        --index-url https://download.pytorch.org/whl/cu118

# Install transformers ecosystem (GPU-ready builds)
RUN /opt/venv/bin/pip install --no-cache-dir \
        transformers==4.39.3 \
        peft==0.10.0 \
        accelerate==0.28.0 \
        bitsandbytes==0.43.1 \
        huggingface_hub>=0.21.0

# -----------------------------------------------------------------------------
# Stage 2: Runtime
# Slim runtime image with CUDA libraries, running as non-root.
# -----------------------------------------------------------------------------
FROM nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 AS runtime

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONDONTWRITEBYTECODE=1 \
    PYTHONUNBUFFERED=1 \
    PIP_NO_CACHE_DIR=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    TRANSFORMERS_CACHE=/model/.cache \
    HF_HOME=/model/.cache \
    CUDA_VISIBLE_DEVICES=0 \
    PORT=8000 \
    HOST=0.0.0.0

WORKDIR /app

# Install runtime Python + basic utils (no compilers needed here)
RUN apt-get update && apt-get install -y --no-install-recommends \
        python3.11 \
        python3.11-venv \
        python3-pip \
        curl \
        git \
    && rm -rf /var/lib/apt/lists/* \
    && ln -sf python3.11 /usr/bin/python

# Copy virtualenv from builder
COPY --from=builder /opt/venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

# Create non-root user for security
ARG UID=1000
ARG GID=1000
RUN groupadd --gid $GID stack && useradd --uid $UID --gid $GID --shell /bin/bash --create-home stack

# Create model mount point
RUN mkdir -p /model && chown stack:stack /model

# Copy inference entrypoint
COPY --chown=stack:stack inference_api.py .

# Switch to non-root
USER stack:stack

# Healthcheck — confirm CUDA libraries are visible
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -sf http://localhost:${PORT}/health || exit 1

EXPOSE ${PORT}

# Model is expected to be mounted at /model at runtime.
# Example: docker run -v /path/to/base_model_qwen7b:/model:ro stack-2.9-gpu
ENV MODEL_PATH=/model

ENTRYPOINT ["python", "inference_api.py"]