Unable to docker build tao_tf1

Running on Linux Ubuntu 22.04.4 LTS with NVIDIA RTX A4000.

Hi, I’m trying to build the tao docker from the tensorflow 1 backend.
After pulling the repo and executing source scripts/envsetup.sh, then executing tao_tf, the docker starts to build but ends because it is unable to install all requirements:

235.0 Successfully built addict future nvidia_dali_tf_plugin_cuda110 mpi4py posix_ipc prettytable pycuda retrying seaborn semver shapely fire antlr4-python3-runtime promise pathtools
235.0 Failed to build h5py pycocotools-fix
235.0 ERROR: Could not build wheels for h5py, pycocotools-fix, which is required to install pyproject.toml-based projects
------
Dockerfile:58
--------------------
  57 |     COPY docker/requirements-pip.txt /workspace/
  58 | >>> RUN pip install -r requirements-pip.txt --use-deprecated=legacy-resolver && \
  59 | >>>     rm -rf requirements-pip.txt
  60 |     
--------------------
ERROR: failed to solve: process "/bin/sh -c pip install -r requirements-pip.txt --use-deprecated=legacy-resolver &&     rm -rf requirements-pip.txt" did not complete successfully: exit code: 1

You can directly pull the tf1 docker from TAO Toolkit | NVIDIA NGC
$ docker pull nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf1.15.5

The goal here is to update and modify some of the code and to rebuild the docker with my updates.

OK, there is another quicker way.
You can login the docker, then vim the code.
For example,
$vim /usr/local/lib/python3.8/dist-packages/nvidia_tao_tf1/cv/common/export/trt_utils.py

Then, you can docker commit a new docker.

Making changes with vim and building a new docker seems overkill and not convenient.
Can you help me building the docker from the source code instead (cf. first post of this topic)?

OK, I will try on my side.

Any updates on this?

Please add below lines between RUN pip install nvidia-pyindex and COPY docker/requirements-pip.txt /workspace/. It works on my side. Verify on an Ubuntu20.04 machine.
$ vim tao_tensorflow1_backend/docker/Dockerfile at main · NVIDIA/tao_tensorflow1_backend · GitHub

RUN pip install Cython==0.29.36
RUN apt install -y libopenmpi-dev
RUN pip install mpi4py

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.