Add gpu guard for broadcast_coalesce#5655
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This patch fixes a bug triggered by #5182 when we have multiple layers in the model, and the DDP is run on a single node, with a subset of GPUs each.
For example, as in the test we run 2 processes on a 8 GPU node, both processes are visible to all GPUs. We create the DDP model by
nn.parallel.DistributedDataParallel(model_DDP, device_ids=gpu_subset)where gpu_subset is 0,1,2,3 for process 1, and 4,5,6,7 for process 2.utils::flatten_dense_tensors(chunk.tensors) will actually create a new Tensor which a flatten version of layer weights. Without this patch, this tensor goes to default GPU 0 despite all layers weights for process 2 are on GPU4, this will further error out when broadcast requires the tensor to be on the GPU 4 for process 2.
The gpu guard inside the for loop has nothing to do with the current bug, I thought it's good to add it as a safety guard.
@apaszke