AttributeError: 'MessageFactory' object has no attribute 'GetPrototype'

[sallos-cyber]

Thank you for publishing your solution!

There is one issue though, when I built your Dockerfile with the same gpu and import tensorflow it throws this error:

AttributeError: 'MessageFactory' object has no attribute 'GetPrototype'

I guess you must have encountered that, too?

[sallos-cyber]

Ok, so there is a newer protobuf version now (6.31.0 at the time of writing). I assume that this one hadn't been released when you wrote your Dockerfile (which was so helpful!). If you force pip to use an older version (5.29.4) it works fine. Maybe you want to put that in the Dockerfile.

[marcoherde]

I have the same error (otherwise it works great!) Could you be so kind to share the change you made to the Dockerfile? I cannot see where to change the version of protobuf (is it protoc?) in the Dockerfile.

[Radi-m]

The recent Tensorflow commit fixes this issue. You have to use a build parameter
--action_env=PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=upb.

Make sure you install the protobuf==5.28.3 instead of the latest one.

My Docker file - requires requirements.txt:

# Stage 1: The Builder
# This stage has all the tools needed to compile TensorFlow (CUDA dev kit, Clang, Bazel, etc.)
FROM nvidia/cuda:12.8.1-cudnn-devel-ubuntu24.04 AS builder
ENV DEBIAN_FRONTEND=noninteractive
WORKDIR /workspace

RUN --mount=target=/var/lib/apt/lists,type=cache,id=apt-lists \
    --mount=target=/var/cache/apt,type=cache,id=apt-cache \
    apt-get update && \
    apt-get install -y wget software-properties-common lsb-release git python3 python3-venv python3-pip

RUN --mount=target=/var/lib/apt/lists,type=cache,id=apt-lists \
    --mount=target=/var/cache/apt,type=cache,id=apt-cache \
    wget -O llvm.sh https://apt.llvm.org/llvm.sh && \
    chmod +x llvm.sh && \
    ./llvm.sh 20 all && \
    update-alternatives --install /usr/bin/clang clang /usr/bin/clang-20 100 && \
    rm llvm.sh

RUN git clone --depth 1 https://github.com/tensorflow/tensorflow.git
WORKDIR /workspace/tensorflow

RUN wget https://github.com/bazelbuild/bazelisk/releases/download/v1.20.0/bazelisk-linux-amd64 -O /usr/bin/bazel && \
    chmod +x /usr/bin/bazel

COPY .tf_configure.bazelrc .

RUN --mount=type=cache,target=/root/.cache/bazel,id=bazel-cache \
    bazel build //tensorflow/tools/pip_package:wheel \
        --repo_env=USE_PYWRAP_RULES=1 \
        --repo_env=WHEEL_NAME=tensorflow \
        --config=cuda \
        --config=cuda_wheel \
        --config=opt && \
    mkdir /dist && \
    cp /workspace/tensorflow/bazel-bin/tensorflow/tools/pip_package/wheel_house/*.whl /dist/

# ------------------------------------------------------------------------------------

# Stage 2: The Final Image
FROM nvidia/cuda:12.8.1-cudnn-runtime-ubuntu24.04 AS final

ENV DEBIAN_FRONTEND=noninteractive
WORKDIR /workspace

# Install only Python and necessary runtime libraries
RUN --mount=target=/var/lib/apt/lists,type=cache,id=apt-lists \
    --mount=target=/var/cache/apt,type=cache,id=apt-cache \
    apt-get update && \
    apt-get install -y python3 python3-venv python3-pip && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

# Set up the Python virtual environment
RUN python3 -m venv venv
ENV VIRTUAL_ENV=/workspace/venv
ENV PATH="$VIRTUAL_ENV/bin:$PATH"

# Copy the requirements file and install the EXACT protobuf version FIRST
COPY requirements.txt .
RUN --mount=target=/root/.cache/pip,type=cache,id=pip-global-cache \
    pip install -r requirements.txt

# Now, copy the compiled TensorFlow wheel from the builder stage
COPY --from=builder /dist/*.whl .
# Install the TensorFlow wheel. Pip will see protobuf is already installed and not change it.
RUN --mount=target=/root/.cache/pip,type=cache,id=pip-global-cache \
    pip install *.whl

# Clean up the wheel file after installation
RUN rm *.whl

requirements.txt

# This version must match the gencode version used to compile TensorFlow
protobuf==5.28.3