That's a good point, I can undo the xgboost install step. Would this require installing from source, or the latest stable version (using pip) is good enough?

I think it should be source.

@RAMitchell I've updated the dockerfile but the xgboost install isn't working with the following error:

CMake Error at plugin/CMakeLists.txt:8 (message):
  Could not locate RMM library

I can run the same command within the container, but the Dockerfile for some reason fails. Do you know what could be the reason for this?

I don't I'm afraid. Maybe @hcho3 has an idea?

RMM_ROOT=/opt/conda should be revised to point to the correct location of RMM library in the Conda environment.

@hcho3 is it /rapids/rmm or elsewhere?

However, this worked when I ran it inside rapidsai/rapidsai-dev:21.06-cuda11.0-devel-ubuntu18.04-py3.8

git config --global http.sslVerify false && \
    git clone --recursive https://github.com/dmlc/xgboost /opt/xgboost && \
    cd /opt/xgboost && \
    mkdir build && \
    cd build && \
    RMM_ROOT=/opt/conda cmake .. \
        -DUSE_CUDA=ON \
        -DUSE_NCCL=ON \
        -DPLUGIN_RMM=ON && \
    make -j4 && \
    cd ../python-package && \
    pip uninstall -y xgboost && \
    python setup.py install

Can you run echo $CONDA_PREFIX inside the container? That's where RMM is installed.

Thanks! That helped /opt/conda/envs/rapids