Fix uneven head sequence parallelism bug (#6774) #6797
+1
−1
Conversation
This file contains 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
|
Fix #6774 |
|
LGTM， Many thanks! |
inkcherry
reviewed
Dec 2, 2024
| @@ -155,7 +155,7 @@ def single_all_to_all(input, scatter_idx, gather_idx, batch_dim_idx, group, asyn | |||
| # we only need num_heads once | |||
| num_heads = input.shape[2] | |||
|
|
|||
| if get_num_kv_heads() is not None or num_heads % seq_world_size != 0: | |||
| if get_num_kv_heads() is not None or (num_heads % seq_world_size != 0 and gather_idx < 2): | |||
There was a problem hiding this comment.
By the way, can we use scatter_idx for the judgment here? This way, we can stay consistent with other parts by uniformly using scatter_idx to determine whether the current state is total_head or total_seq_len
There was a problem hiding this comment.
Ditto @inkcherry! Future refactor will have a better naming to differentiate between N/p (pre/post-attention) and h/p (attention block)
There was a problem hiding this comment.
Hi I agree, scatter_idx booleans are confusing. Maybe I can share my opinion on the refactoring on Wed if thats ok.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Here
gather_idx < 2representsis_first_all2all. During the first all2all,uneven_head_all2allwill be called if eithernum_heads % seq_world_size != 0orget_num_kv_heads() is None.During the second all2all, it'll return return
uneven_head_all2allif and only ifget_num_kv_heads() is Nonewhich is always set during the first uneven all2all. This means that there will no longer be issue whereuneven_head_all2allis returned for the second all2all because ofnum_heads % seq_world_size != 0.