# Embedding Service Lite - GPU
# CUDA 12.8 — поддержка современных карт (RTX 50xx и др.)
FROM pytorch/pytorch:2.7.0-cuda12.8-cudnn9-runtime
LABEL maintainer="Roman Zateev"
LABEL description="Embedding Service Lite (GPU) for 1C Help"
WORKDIR /app
# Дополнительные системные зависимости
RUN apt-get update && apt-get install -y \
git \
build-essential \
curl \
libgomp1 \
unzip \
&& rm -rf /var/lib/apt/lists/* \
&& apt-get clean
# Python зависимости (без torch — он уже в базовом образе)
COPY requirements-embedding-lite.txt /tmp/requirements-embedding-lite.txt
RUN pip install --no-cache-dir --upgrade pip && \
pip install --no-cache-dir -r /tmp/requirements-embedding-lite.txt
# Копируем код сервиса
COPY embedding_service_lite/ /app/embedding_service_lite/
# Директории для BM25 и кешей моделей
ENV TORCH_HOME=/cache/torch
ENV HF_HOME=/cache/huggingface
RUN mkdir -p /cache/torch /cache/huggingface /app/data/bm25-storage
ENV EMBEDDING_DEVICE=gpu
ENV PYTHONUNBUFFERED=1
ENV CUDA_DEVICE_ORDER=PCI_BUS_ID
ENV CUDA_VISIBLE_DEVICES=0
EXPOSE 8004
HEALTHCHECK --interval=30s --timeout=30s --start-period=60s --retries=3 \
CMD curl -f http://localhost:8004/health || exit 1
CMD ["uvicorn", "embedding_service_lite.main:app", "--host", "0.0.0.0", "--port", "8004"]
