Dockerfile

FROM pytorch/pytorch:2.6.0-cuda12.6-cudnn9-devel

WORKDIR /app

RUN apt-get update && apt-get install -y git ffmpeg libsndfile1 build-essential && rm -rf /var/lib/apt/lists/*

RUN pip install --no-cache-dir \
    transformers einops loguru ormsgpack hydra-core omegaconf safetensors \
    soundfile datasets huggingface_hub tiktoken loralib pyrootutils natsort \
    librosa rich resampy pydub cachetools silero-vad descript-audio-codec \
    lightning zstandard opencc-python-reimplemented einx torchaudio

RUN git clone --depth 1 https://github.com/fishaudio/fish-speech.git /app/fish-speech && \
    cd /app/fish-speech && pip install --no-deps -e .

COPY gen_samples.py /app/gen_samples.py
COPY reference/ /app/reference/

ENV PYTHONPATH=/app/fish-speech:$PYTHONPATH
ENV TOKENIZERS_PARALLELISM=false
ENV MPLCONFIGDIR=/tmp/matplotlib
ENV HF_HOME=/tmp/hf_cache
ENV XDG_CACHE_HOME=/tmp/cache

CMD ["python", "/app/gen_samples.py"]