simplify embedding + first transformer block TP#314
Merged
Conversation
as titled, we can directly specify the rowwise parallel embedding output layouts be shard on sequence dim, so that we don't need the first layer prepare input. Switching to output_layouts = Shard(1) would also trigger reduce_scatter instead of allreduce for embedding layer, which could give some small perf wins
Contributor
|
cool :) |
tianyu-l
added a commit
to pytorch/examples
that referenced
this pull request
May 16, 2024
…first transformer block" Following changes in pytorch/torchtitan#314, to apply a reduce-scatter instead of the more expensive all-reduce + local chunk. cross PR with pytorch/tutorials#2871 [ghstack-poisoned]
Collaborator
|
Reminder to self: We should update the comment and remove the |
tianyu-l
pushed a commit
to tianyu-l/torchtitan_intern24
that referenced
this pull request
Aug 16, 2024
as titled, we can directly specify the rowwise parallel embedding output layouts be shard on sequence dim, so that we don't need the first layer prepare input. Switching to output_layouts = Shard(1) would also trigger reduce_scatter instead of allreduce for embedding layer, which could give some small perf wins
philippguevorguian
pushed a commit
to YerevaNN/YNNtitan
that referenced
this pull request
Aug 17, 2024
as titled, we can directly specify the rowwise parallel embedding output layouts be shard on sequence dim, so that we don't need the first layer prepare input. Switching to output_layouts = Shard(1) would also trigger reduce_scatter instead of allreduce for embedding layer, which could give some small perf wins
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
5 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.
as titled, we can directly specify the rowwise parallel embedding output layouts be shard on sequence dim, so that we don't need the first layer prepare input.
Switching to output_layouts = Shard(1) would also trigger reduce_scatter instead of allreduce for embedding layer, which could give some small perf wins