Dockerfile.tts_sdp

FROM pytorch/pytorch:2.4.1-cuda12.1-cudnn9-devel

ARG DEBIAN_FRONTEND=noninteractive

ENV TZ=America/Los_Angeles

# Install basics
RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential \
    bzip2 \
    ca-certificates \
    libsox-fmt-mp3 \
    cmake \
    curl \
    ffmpeg \
    g++ \
    sox \
    unzip \
    vim \
    wget

# Update pip
RUN pip install --upgrade pip

# Link all cudnn .so libraries for runtime
RUN ln -s /opt/conda/lib/python3.11/site-packages/nvidia/cudnn/include/cudnn*.h /usr/include/
RUN mkdir -p /usr/local/cuda/lib64
RUN ln -s /opt/conda/lib/python3.11/site-packages/nvidia/cudnn/lib/libcudnn*.so* /usr/local/cuda/lib64/
ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH


# Copy NeMo SDP
WORKDIR /src
COPY . /src/NeMo-speech-data-processor
RUN rm -rf /src/NeMo-speech-data-processor/.git

# Install requirements
WORKDIR /src/NeMo-speech-data-processor
RUN pip install -r requirements/main.txt 
RUN pip install -r requirements/tts.txt
RUN pip install flash-attn --no-build-isolation
RUN pip install https://github.com/LahiLuk/YouTokenToMe/archive/master.zip
RUN pip install --no-build-isolation megatron-core transformer_engine[pytorch]==2.4.0
RUN pip install nemo_toolkit['all']==2.3.2

WORKDIR /src/NeMo-speech-data-processor