feat: Introduce Gitea CI/CD workflows, refactor Docker deployment with dedicated dev/prod compose files and enhanced ROCm GPU support, and update RAG documentation for new AI and ASR infrastructure.

This commit is contained in:
2026-01-24 14:50:19 +00:00
parent 09d70af58f
commit acfba090e4
29 changed files with 2676 additions and 918 deletions

View File

@@ -1,16 +1,4 @@
# syntax=docker/dockerfile:1
# NoteFlow ROCm Docker Image
# For AMD GPU support using PyTorch ROCm
#
# Build:
# docker build -f docker/Dockerfile.rocm -t noteflow:rocm .
#
# Run (with GPU access):
# docker run --device=/dev/kfd --device=/dev/dri --group-add video --group-add render \
# --security-opt seccomp=unconfined \
# -v /path/to/models:/workspace/models \
# noteflow:rocm
ARG ROCM_VERSION=6.4.1
ARG ROCM_PYTORCH_RELEASE=2.6.0
ARG SPACY_MODEL_URL=https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
@@ -29,10 +17,8 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
UV_LINK_MODE=copy \
PATH=/usr/local/bin:$PATH
# Install uv
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
# Install system dependencies
RUN apt-get update && apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
@@ -43,72 +29,51 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
git \
&& rm -rf /var/lib/apt/lists/*
# =============================================================================
# Server Stage - ROCm
# =============================================================================
FROM base AS server
WORKDIR /workspace
# Copy dependency files first for better layer caching
COPY pyproject.toml uv.lock* ./
COPY README.md ./
COPY src ./src/
# Create venv with access to system site-packages (ROCm torch)
ENV VIRTUAL_ENV=/opt/venv
RUN uv venv --system-site-packages ${VIRTUAL_ENV}
ENV PATH="${VIRTUAL_ENV}/bin:$PATH"
# Install NoteFlow with ROCm extras (into venv)
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python -e ".[rocm,optional]"
# Improve redis client performance and silence hiredis warning.
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python hiredis
# Install spaCy small English model for NER (baked into image)
# =============================================================================
# Production Server Stage (minimal deps)
# =============================================================================
FROM base AS server
ARG SPACY_MODEL_URL
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python -e ".[rocm]"
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python ${SPACY_MODEL_URL}
# Copy remaining files (scripts, configs, etc.)
COPY . .
# Environment variables for ROCm
ENV ROCM_PATH=/opt/rocm \
HIP_VISIBLE_DEVICES=0 \
HSA_OVERRIDE_GFX_VERSION="" \
NOTEFLOW_ASR_DEVICE=rocm \
NOTEFLOW_FEATURE_ROCM_ENABLED=true
# gRPC server port
RUN useradd --create-home --shell /bin/bash noteflow
USER noteflow
EXPOSE 50051
# Health check
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
CMD python -c "import grpc; channel = grpc.insecure_channel('localhost:50051'); grpc.channel_ready_future(channel).result(timeout=5)" || exit 1
# Run gRPC server
CMD ["python", "-m", "noteflow.grpc.server"]
# =============================================================================
# Server Dev Stage - ROCm (hot reload)
# Development Server Stage (all extras + hot reload)
# =============================================================================
FROM base AS server-dev
WORKDIR /workspace
COPY pyproject.toml uv.lock* ./
COPY README.md ./
COPY src ./src/
ENV VIRTUAL_ENV=/opt/venv
RUN uv venv --system-site-packages ${VIRTUAL_ENV}
ENV PATH="${VIRTUAL_ENV}/bin:$PATH"
FROM base AS dev
ARG SPACY_MODEL_URL
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python -e ".[rocm,optional]"
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python hiredis
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python watchfiles
ARG SPACY_MODEL_URL
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python ${SPACY_MODEL_URL}
COPY . .
@@ -122,38 +87,3 @@ ENV ROCM_PATH=/opt/rocm \
EXPOSE 50051
CMD ["python", "scripts/dev_watch_server.py"]
# =============================================================================
# Server Full Stage - ROCm (optional extras)
# =============================================================================
FROM base AS server-full
WORKDIR /workspace
COPY pyproject.toml uv.lock* ./
COPY README.md ./
COPY src ./src/
ENV VIRTUAL_ENV=/opt/venv
RUN uv venv --system-site-packages ${VIRTUAL_ENV}
ENV PATH="${VIRTUAL_ENV}/bin:$PATH"
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python -e ".[rocm,optional]"
ARG SPACY_MODEL_URL
RUN uv pip install --python ${VIRTUAL_ENV}/bin/python ${SPACY_MODEL_URL}
COPY . .
ENV ROCM_PATH=/opt/rocm \
HIP_VISIBLE_DEVICES=0 \
HSA_OVERRIDE_GFX_VERSION="" \
NOTEFLOW_ASR_DEVICE=rocm \
NOTEFLOW_FEATURE_ROCM_ENABLED=true
EXPOSE 50051
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
CMD python -c "import grpc; channel = grpc.insecure_channel('localhost:50051'); grpc.channel_ready_future(channel).result(timeout=5)" || exit 1
CMD ["python", "-m", "noteflow.grpc.server"]

View File

@@ -1,21 +1,8 @@
#!/bin/bash
# GPU entrypoint script
# Sets LD_LIBRARY_PATH to prioritize PyTorch's bundled cuDNN 9.8.0 over system cuDNN 9.1.0
set -e
# PyTorch bundles cuDNN 9.8.0 in its site-packages
# We must add these paths FIRST to override system cuDNN 9.1.0
PYTORCH_NVIDIA_LIBS="/workspace/.venv/lib/python3.12/site-packages/nvidia"
export LD_LIBRARY_PATH="${PYTORCH_NVIDIA_LIBS}/cudnn/lib:${PYTORCH_NVIDIA_LIBS}/cublas/lib:${PYTORCH_NVIDIA_LIBS}/cuda_runtime/lib:${PYTORCH_NVIDIA_LIBS}/cufft/lib:${PYTORCH_NVIDIA_LIBS}/cusolver/lib:${PYTORCH_NVIDIA_LIBS}/cusparse/lib:${PYTORCH_NVIDIA_LIBS}/nccl/lib:${PYTORCH_NVIDIA_LIBS}/nvtx/lib:/usr/local/cuda/lib64"
echo "=== GPU Entrypoint ==="
echo "LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
echo "Checking cuDNN libraries..."
ls -la "${PYTORCH_NVIDIA_LIBS}/cudnn/lib/" 2>/dev/null | head -5 || echo "cuDNN libs not found (will be installed on first run)"
echo "======================"
# Run uv sync to ensure dependencies are installed
uv sync --frozen --group dev --all-extras
# Execute the command passed to docker run
exec "$@"

View File

@@ -1,35 +1,26 @@
# syntax=docker/dockerfile:1
# GPU-enabled server Dockerfile with CUDA support
# Use this for systems with NVIDIA GPUs
ARG PYTHON_VERSION=3.12
ARG CUDA_VERSION=12.4.1
ARG SPACY_MODEL_URL=https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
# =============================================================================
# Python Stage - Get Python 3.12 from official image
# =============================================================================
FROM python:3.12-slim-bookworm AS python-base
FROM python:${PYTHON_VERSION}-slim-bookworm AS python-base
# =============================================================================
# Base Stage - NVIDIA CUDA with cuDNN for GPU-accelerated inference
# =============================================================================
# Using NVIDIA's official CUDA image with cuDNN 9.x for CTranslate2/faster-whisper
# The runtime image includes cuDNN libraries required for GPU inference
FROM nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04 AS base
FROM nvidia/cuda:${CUDA_VERSION}-cudnn-runtime-ubuntu22.04 AS base
# CUDA/cuDNN environment variables
# NOTE: PyTorch bundles cuDNN 9.8.0, but system has 9.1.0
# We set LD_LIBRARY_PATH at runtime to prioritize PyTorch's bundled cuDNN
ENV PYTHONDONTWRITEBYTECODE=1 \
PYTHONUNBUFFERED=1 \
UV_COMPILE_BYTECODE=1 \
UV_LINK_MODE=copy \
# CUDA environment - these tell nvidia-container-runtime to inject GPU
NVIDIA_VISIBLE_DEVICES=all \
NVIDIA_DRIVER_CAPABILITIES=compute,utility \
# Base CUDA path (cuDNN paths added at runtime to use PyTorch's bundled version)
LD_LIBRARY_PATH=/usr/local/cuda/lib64 \
# Python path configuration
PATH=/usr/local/bin:$PATH
# Copy Python 3.12 from official image (avoids PPA network issues)
COPY --from=python-base /usr/local/bin/python3.12 /usr/local/bin/python3.12
COPY --from=python-base /usr/local/bin/python3 /usr/local/bin/python3
COPY --from=python-base /usr/local/bin/pip3 /usr/local/bin/pip3
@@ -37,94 +28,92 @@ COPY --from=python-base /usr/local/lib/python3.12 /usr/local/lib/python3.12
COPY --from=python-base /usr/local/include/python3.12 /usr/local/include/python3.12
COPY --from=python-base /usr/local/lib/libpython3.12.so* /usr/local/lib/
# Create symlinks for python/pip commands
RUN ln -sf /usr/local/bin/python3.12 /usr/local/bin/python \
&& ln -sf /usr/local/bin/pip3 /usr/local/bin/pip \
&& ldconfig
# Install uv and system dependencies
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
# Core build/runtime deps for project packages (sounddevice, asyncpg, cryptography).
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
libportaudio2 \
libsndfile1 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /workspace
# Copy dependency files first for better layer caching
COPY pyproject.toml uv.lock* ./
# =============================================================================
# Server Stage - GPU Enabled
# Production Server Stage (minimal deps)
# =============================================================================
FROM base AS build
ARG SPACY_MODEL_URL
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
&& rm -rf /var/lib/apt/lists/*
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --no-dev
COPY . .
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-dev
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install "${SPACY_MODEL_URL}"
FROM base AS server
# Install dependencies with CUDA-enabled PyTorch
# The --extra-index-url ensures we get CUDA-enabled torch
COPY --from=build /workspace /workspace
COPY docker/entrypoint-gpu.sh /usr/local/bin/entrypoint-gpu.sh
RUN chmod +x /usr/local/bin/entrypoint-gpu.sh
RUN useradd --create-home --shell /bin/bash noteflow
USER noteflow
EXPOSE 50051
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
CMD python -c "import grpc; channel = grpc.insecure_channel('localhost:50051'); grpc.channel_ready_future(channel).result(timeout=5)" || exit 1
ENTRYPOINT ["/usr/local/bin/entrypoint-gpu.sh"]
CMD ["uv", "run", "python", "-m", "noteflow.grpc.server"]
# =============================================================================
# Development Server Stage (all extras + dev deps + hot reload)
# =============================================================================
FROM base AS dev
ARG SPACY_MODEL_URL
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
&& rm -rf /var/lib/apt/lists/*
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --group dev --all-extras
# Copy source code
COPY . .
# Install the project itself
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --group dev --all-extras
# Install spaCy small English model for NER (baked into image)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
uv pip install "${SPACY_MODEL_URL}"
# Verify CUDA and cuDNN are accessible
# Note: torch.cuda.is_available() may return False during build (no GPU)
# but cuDNN libraries should be present in the image for runtime
RUN python -c "import torch; print(f'PyTorch: {torch.__version__}'); print(f'CUDA available: {torch.cuda.is_available()}'); print(f'cuDNN version: {torch.backends.cudnn.version() if torch.backends.cudnn.is_available() else \"N/A\"}')" || true
# Verify cuDNN shared libraries are present
RUN ldconfig -p | grep -i cudnn || echo "cuDNN libraries will be available at runtime"
# Copy GPU entrypoint script that sets LD_LIBRARY_PATH correctly
COPY docker/entrypoint-gpu.sh /usr/local/bin/entrypoint-gpu.sh
RUN chmod +x /usr/local/bin/entrypoint-gpu.sh
EXPOSE 50051
# Use entrypoint script to set LD_LIBRARY_PATH correctly at runtime
# This ensures PyTorch's bundled cuDNN 9.8.0 takes priority over system cuDNN 9.1.0
ENTRYPOINT ["/usr/local/bin/entrypoint-gpu.sh"]
CMD ["uv", "run", "python", "scripts/dev_watch_server.py"]
# =============================================================================
# Server Production Stage - GPU Enabled with all extras
# =============================================================================
FROM base AS server-full
# Install all dependencies including optional extras
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --group dev --all-extras
# Copy source code
COPY . .
# Install the project itself
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --group dev --all-extras
# Install spaCy small English model for NER (baked into image)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
# Copy GPU entrypoint script that sets LD_LIBRARY_PATH correctly
COPY docker/entrypoint-gpu.sh /usr/local/bin/entrypoint-gpu.sh
RUN chmod +x /usr/local/bin/entrypoint-gpu.sh
EXPOSE 50051
# Use entrypoint script to set LD_LIBRARY_PATH correctly at runtime
# This ensures PyTorch's bundled cuDNN 9.8.0 takes priority over system cuDNN 9.1.0
ENTRYPOINT ["/usr/local/bin/entrypoint-gpu.sh"]
CMD ["uv", "run", "python", "scripts/dev_watch_server.py"]

View File

@@ -1,91 +1,88 @@
# syntax=docker/dockerfile:1
FROM python:3.12-bookworm AS base
ARG PYTHON_VERSION=3.12
ARG SPACY_MODEL_URL=https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
FROM python:${PYTHON_VERSION}-slim-bookworm AS base
ENV PYTHONDONTWRITEBYTECODE=1 \
PYTHONUNBUFFERED=1 \
UV_COMPILE_BYTECODE=1 \
UV_LINK_MODE=copy
# Install uv
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
# Core build/runtime deps for project packages (sounddevice, asyncpg, cryptography).
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
libportaudio2 \
libsndfile1 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /workspace
# Copy dependency files first for better layer caching
COPY pyproject.toml uv.lock* ./
# =============================================================================
# Server Stage
# Production Server Stage (minimal deps, no dev tooling)
# =============================================================================
FROM base AS build
ARG SPACY_MODEL_URL
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
&& rm -rf /var/lib/apt/lists/*
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --no-dev
COPY . .
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-dev
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install "${SPACY_MODEL_URL}"
FROM base AS server
# Install dependencies (server needs dev deps for watchfiles)
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --group dev --all-extras
COPY --from=build /workspace /workspace
# Copy source code
COPY . .
# Install the project itself
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --group dev --all-extras
# Install spaCy small English model for NER (baked into image)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
RUN useradd --create-home --shell /bin/bash noteflow
USER noteflow
EXPOSE 50051
CMD ["sh", "-c", "uv sync --frozen --group dev --all-extras && uv run python scripts/dev_watch_server.py"]
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
CMD python -c "import grpc; channel = grpc.insecure_channel('localhost:50051'); grpc.channel_ready_future(channel).result(timeout=5)" || exit 1
CMD ["uv", "run", "python", "-m", "noteflow.grpc.server"]
# =============================================================================
# Server Production Stage (all optional dependencies)
# Development Server Stage (all extras + dev deps + hot reload)
# =============================================================================
FROM base AS server-full
# Install all dependencies including optional extras
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --group dev --all-extras
# Copy source code
COPY . .
# Install the project itself
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --group dev --all-extras
# Install spaCy small English model for NER (baked into image)
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
EXPOSE 50051
CMD ["sh", "-c", "uv sync --frozen --group dev --all-extras && uv run python scripts/dev_watch_server.py"]
# -----------------------------------------------------------------------------
# NER stage: Add spaCy model for named entity recognition
# -----------------------------------------------------------------------------
FROM base AS with-ner
# Install NER dependencies and download spaCy model
RUN uv pip install -e ".[ner]" \
&& uv run python -m spacy download en_core_web_sm
# Verify model is available
RUN uv run python -c "import spacy; nlp = spacy.load('en_core_web_sm'); print('NER model loaded successfully')"
# -----------------------------------------------------------------------------
# Development target (default)
# -----------------------------------------------------------------------------
FROM base AS dev
ARG SPACY_MODEL_URL
CMD ["sh", "-c", "uv sync --frozen --group dev --all-extras && uv run python scripts/dev_watch_server.py"]
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential \
pkg-config \
portaudio19-dev \
&& rm -rf /var/lib/apt/lists/*
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --no-install-project --group dev --all-extras
COPY . .
RUN --mount=type=cache,target=/root/.cache/uv \
uv sync --frozen --group dev --all-extras
RUN --mount=type=cache,target=/root/.cache/uv \
uv pip install "${SPACY_MODEL_URL}"
EXPOSE 50051
CMD ["uv", "run", "python", "scripts/dev_watch_server.py"]