Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
The combination of PyTorch 1.9 and the HOROVOD_GPU_ALLREDUCE=NCCL option (which means to use NCCL specifically for the All-Reduce algorithm only) has been resulting in extra processes on the GPUs and consequent memory errors. Configuring NCCL to be used for all operations appears to eliminate this problem and we probably should have been doing that anyway. Also, working around a recent regression in Pillow by just pinning to an older release (python-pillow/Pillow#5571).
See also: determined-ai/environments#110
Test Plan
https://app.circleci.com/pipelines/github/determined-ai/determined?branch=all_operations_nccl
At the time of this writing, one job is still running but the rest have passed.