Dockerfile

FROM nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu22.04

WORKDIR /app

ENV DEBIAN_FRONTEND=noninteractive \
    PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1

# system deps
RUN apt-get update && apt-get install -y \
    python3.10 python3.10-venv python3-pip \
    ffmpeg git libsndfile1 curl tzdata wget build-essential \
    && ln -sf /usr/bin/python3.10 /usr/bin/python \
    && ln -sf /usr/bin/pip3 /usr/bin/pip \
    && rm -rf /var/lib/apt/lists/*

# timezone
RUN ln -fs /usr/share/zoneinfo/Asia/Bangkok /etc/localtime && \
    dpkg-reconfigure -f noninteractive tzdata

# caches
RUN mkdir -p /tmp/hf_cache /tmp/torch_cache /tmp/matplotlib /tmp/xdg_cache /tmp/home /tmp/uploads /tmp/pythainlp_data \
    && chmod -R 777 /tmp/hf_cache /tmp/torch_cache /tmp/matplotlib /tmp/xdg_cache /tmp/home /tmp/uploads /tmp/pythainlp_data

# --------- IMPORTANT: limit threads to avoid libgomp spawn fail ---------
ENV OMP_NUM_THREADS=1 \
    OPENBLAS_NUM_THREADS=1 \
    MKL_NUM_THREADS=1 \
    NUMEXPR_NUM_THREADS=1 \
    VECLIB_MAXIMUM_THREADS=1 \
    BLIS_NUM_THREADS=1 \
    TOKENIZERS_PARALLELISM=false \
    PL_DISABLE_VERSION_CHECK=1 \
    HUGGINGFACE_HUB_CACHE=/tmp/hf_cache \
    HF_HOME=/tmp/hf_cache \
    HF_CACHE=/tmp/hf_cache \
    TRANSFORMERS_CACHE=/tmp/hf_cache \
    TORCH_HOME=/tmp/torch_cache \
    MPLCONFIGDIR=/tmp/matplotlib \
    XDG_CACHE_HOME=/tmp/xdg_cache \
    HOME=/tmp/home \
    TMPDIR=/tmp \
    TEMP=/tmp \
    TMP=/tmp \
    LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH \
    PYTHAINLP_DATA_DIR=/tmp/pythainlp_data

# python deps
COPY requirements.txt .
RUN pip install --upgrade pip

# torch cuda12.1
RUN pip install --no-cache-dir \
    torch==2.3.0+cu121 torchvision==0.18.0+cu121 torchaudio==2.3.0 \
    --extra-index-url https://download.pytorch.org/whl/cu121

# others
RUN pip install --no-cache-dir -r requirements.txt

# app
COPY . /app

EXPOSE 7860

# run: limit workers & concurrency; set ulimit before start
CMD ["uvicorn","app:app","--host","0.0.0.0","--port","7860","--workers","1","--limit-concurrency","8","--timeout-keep-alive","120","--timeout-graceful-shutdown","120"]