Dockerfile.rdna4

FROM rocm/dev-ubuntu-22.04:latest

ENV PYTHONDONTWRITEBYTECODE=1
ENV PYTHONUNBUFFERED=1
ENV DEBIAN_FRONTEND=noninteractive
ENV HF_HOME=/app/hf_cache

RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    libsndfile1 \
    ffmpeg \
    python3 \
    python3-pip \
    python3-dev \
    python3-venv \
    git \
    rocm-ml-libraries \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

RUN ln -s /usr/bin/python3 /usr/bin/python

WORKDIR /app

# Self-contained RDNA4 requirements (no dependency on other requirements files)
COPY requirements-rdna4-init.txt ./requirements-rdna4-init.txt
COPY requirements-rdna4.txt ./requirements-rdna4.txt

# Install order mirrors Dockerfile.rocm: ROCm torch first, then app deps,
# then chatterbox with --no-deps to prevent pip from replacing ROCm torch
RUN python3 -m pip install --no-cache-dir --upgrade pip && \
    python3 -m pip install --no-cache-dir -r requirements-rdna4-init.txt && \
    python3 -m pip install --no-cache-dir -r requirements-rdna4.txt && \
    python3 -m pip install --no-cache-dir --no-deps git+https://github.com/devnen/chatterbox-v2.git@master s3tokenizer==0.3.0 onnx==1.16.0 && \
    python3 -m pip install --no-cache-dir "protobuf>=4.25.0"

# Patches for torch 2.9+ compatibility (required when using ROCm 7.2 wheels):
# s3tokenizer: numpy defaults to float64; cast to float32 to avoid
# float64 STFT → float64 mel → assertion failure
RUN sed -i \
    's/wav = torch\.from_numpy(wav)$/wav = torch.from_numpy(wav).float()/' \
    /usr/local/lib/python3.10/dist-packages/chatterbox/models/s3tokenizer/s3tokenizer.py && \
    sed -i \
    's/mel_spec = self\._mel_filters\.to(self\.device) @ magnitudes/mel_spec = self._mel_filters.to(self.device).to(magnitudes.dtype) @ magnitudes/' \
    /usr/local/lib/python3.10/dist-packages/chatterbox/models/s3tokenizer/s3tokenizer.py

# voice_encoder: melspectrogram() returns float64 → LSTM requires float32
RUN sed -i \
    's/utt_embeds = self\.inference(mels\.to(self\.device),/utt_embeds = self.inference(mels.to(self.device).float(),/' \
    /usr/local/lib/python3.10/dist-packages/chatterbox/models/voice_encoder/voice_encoder.py

COPY . .
RUN mkdir -p model_cache reference_audio outputs voices logs hf_cache

EXPOSE 8004
CMD ["python3", "server.py"]