Please include mamba/libmamba in your base and studiolab environment

[taureandyernv]

Is your feature request related to a problem? Please describe. Conda take as while to solve. Mamba/libmamba is faster and is recommended by Anaconda: https://www.anaconda.com/blog/a-faster-conda-for-a-growing-community. May you please add it to the base environment, so that we can spend more time actually using the GPU than waiting for conda to solve :) Either would be good: https://anaconda.org/conda-forge/mamba

Describe the solution you'd like Add mamba to SageMaker Studio Lab base so that you can mamba install a new custom environment quickly

Describe alternatives you've considered

• installing Mamba first, then conda installing
• pip install
• Conda install

[icoxfog417]

Thank you for the proposal. Definitely mamba solves the waiting time to prepare the environment. I will take the comparison between conda and mamba for major frameworks such as PyTorch to clarify the benefit. We are glad if you share some benchmarks that compare the time of conda and mamba.

[tcmaps]

<snake> install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia -c conda-forge -y

real    3m5.016s
user    2m53.217s
sys     0m3.843s
(conda) studio-lab-user@default:~$

real    0m9.198s
user    0m16.509s
sys     0m1.793s
(mamba) studio-lab-user@default:~$

@icoxfog417 ran twice on new envs to cache pkgs and only consider solve & link time

[rsignell]

@icoxfog417 I returned to AWS sagemaker studio lab after a long time away and disappointed to find it still uses conda as a solver instead of mamba. Anaconda itself is suggesting to use libmamba!.

If you can make this happen it would really make a much better experience for Studio Lab users!

[icoxfog417]

@tcmaps @rsignell So sorry for the wait on this. We've already acknowledged the benefit of new conda version and are hustling to update the base conda version. Let me confirm the progress.