Dockerfile

FROM lmsysorg/sglang:v0.5.8-cu129-amd64-runtime@sha256:85feb856fb76c986cc31001b438d9d243c805378a9e63254fdaaf6cb1740d332

USER root

RUN apt-get update && \
    apt-get install -y --no-install-recommends curl git libsndfile1 ninja-build && \
    rm -rf /var/lib/apt/lists/* && \
    mkdir -p /home/user/app /models /opt/sglang-omni /opt/audio8-tts \
      /tmp/audio8-flashinfer /tmp/audio8-torchinductor && \
    chown -R 1000:1000 /home/user /models \
      /tmp/audio8-flashinfer /tmp/audio8-torchinductor

WORKDIR /home/user/app

ARG SGLANG_OMNI_REVISION=68a572348837f7b004857b4b07993c20ade4c017
ARG AUDIO8_TTS_REVISION=59ca6d27d441302eaa568bcb140b63c50a879859

COPY patches/disable_streaming.patch /tmp/audio8-disable-streaming.patch

RUN git init /opt/sglang-omni && \
    git -C /opt/sglang-omni remote add origin https://github.com/sgl-project/sglang-omni.git && \
    git -C /opt/sglang-omni fetch --depth 1 origin "${SGLANG_OMNI_REVISION}" && \
    git -C /opt/sglang-omni checkout --detach FETCH_HEAD && \
    git init /opt/audio8-tts && \
    git -C /opt/audio8-tts remote add origin https://github.com/Audio8-AI/Audio8_TTS.git && \
    git -C /opt/audio8-tts fetch --depth 1 origin "${AUDIO8_TTS_REVISION}" && \
    git -C /opt/audio8-tts checkout --detach FETCH_HEAD && \
    git -C /opt/audio8-tts apply /tmp/audio8-disable-streaming.patch && \
    rm -rf /opt/sglang-omni/.git /opt/audio8-tts/.git

RUN pip install --no-cache-dir \
      "accelerate==1.14.0" \
      "av==17.1.0" \
      "fastapi==0.139.0" \
      "huggingface-hub==0.36.2" \
      "httpx==0.28.1" \
      "librosa==0.11.0" \
      "msgpack==1.2.1" \
      "numba==0.63.1" \
      "python-multipart==0.0.32" \
      "qwen-vl-utils==0.0.11" \
      "soundfile==0.13.1" \
      "transformers==4.57.1" \
      "typer==0.26.8" \
      "uvicorn==0.49.0" \
      "xxhash==3.8.0" \
      "pyzmq==27.1.0"

ENV PYTHONPATH=/opt/sglang-omni

RUN /opt/audio8-tts/sglang_omni/scripts/install_adapter.sh /opt/sglang-omni

COPY --chown=1000:1000 app.py ./app.py
COPY --chown=1000:1000 static ./static
COPY --chown=1000:1000 examples ./examples
COPY --chown=1000:1000 assets ./assets
COPY --chown=1000:1000 scripts ./scripts

RUN chmod +x ./scripts/*.sh

USER 1000

ARG MODEL_REPO=Audio8/Audio8-TTS-Preview-0.6b
ARG MODEL_REVISION=1b17c91db5f4dccb6914aa4aa5cb0e56661a6c17
RUN python3 -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='${MODEL_REPO}', revision='${MODEL_REVISION}', local_dir='/models/audio8-tts-0.6b')"

ENV HOME=/home/user \
    HF_HOME=/home/user/.cache/huggingface \
    MODEL_PATH=/models/audio8-tts-0.6b \
    PORT=7860 \
    BACKEND_HOST=127.0.0.1 \
    BACKEND_PORT=8010 \
    AUDIO8_TTS_MEM_FRACTION_STATIC=0.60 \
    AUDIO8_TTS_MAX_RUNNING_REQUESTS=1 \
    AUDIO8_TTS_MAX_TEXT_UNITS=150 \
    AUDIO8_TTS_MAX_RAW_TEXT_CHARS=1000 \
    AUDIO8_TTS_MAX_NEW_TOKENS=1024 \
    UI_MAX_CONCURRENCY=1 \
    AUDIO8_TTS_DISABLE_CUDA_GRAPH=0 \
    AUDIO8_TTS_ENABLE_TORCH_COMPILE=1 \
    AUDIO8_TTS_VERIFY_FAST_KV=1 \
    FLASHINFER_WORKSPACE_BASE=/tmp/audio8-flashinfer \
    TORCHINDUCTOR_CACHE_DIR=/tmp/audio8-torchinductor \
    PYTHONUNBUFFERED=1

EXPOSE 7860

CMD ["/home/user/app/scripts/start.sh"]