Implement gather before attn #6378
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
ch-wan
requested review from
ByronHsu,
Ying1123,
hnyls2002,
ispobock,
merrymercy,
zhaochenyang20 and
zhyncs
as code owners
fzyzcjy
reviewed
May 27, 2025
| self._communicate_simple_fn = CommunicateSimpleFn.get_fn( | ||
| input_mode=self.layer_scatter_modes.layer_input_mode, | ||
| output_mode=self.layer_scatter_modes.attn_mode, | ||
| hidden_states_input_mode=self.layer_scatter_modes.attn_mode, |
There was a problem hiding this comment.
wondering whether it should be
Suggested change
| hidden_states_input_mode=self.layer_scatter_modes.attn_mode, | |
| hidden_states_input_mode=self.layer_scatter_modes.layer_input_mode, |
| output_mode=self.layer_scatter_modes.attn_mode, | ||
| hidden_states_input_mode=self.layer_scatter_modes.attn_mode, | ||
| residual_input_mode=self.layer_scatter_modes.layer_input_mode, | ||
| hidden_states_output_mode=self.layer_scatter_modes.mlp_mode, |
There was a problem hiding this comment.
Suggested change
| hidden_states_output_mode=self.layer_scatter_modes.mlp_mode, | |
| hidden_states_output_mode=self.layer_scatter_modes.attn_mode, |
| hidden_states_input_mode=self.layer_scatter_modes.attn_mode, | ||
| residual_input_mode=self.layer_scatter_modes.layer_input_mode, | ||
| hidden_states_output_mode=self.layer_scatter_modes.mlp_mode, | ||
| residual_output_mode=self.layer_scatter_modes.middle_residual_mode, |
There was a problem hiding this comment.
if it is like that, then maybe _communicate_with_all_reduce_and_layer_norm_fn's residual residual mode should be changed
Comment on lines
283
to
288
| if ( | ||
| hidden_states_input_mode == ScatterMode.TP_ATTN_FULL | ||
| and residual_input_mode == ScatterMode.SCATTERED | ||
| and hidden_states_output_mode == ScatterMode.TP_ATTN_FULL | ||
| and residual_output_mode == ScatterMode.SCATTERED | ||
| ): |
There was a problem hiding this comment.
hmm, this branch looks like "input === output", then we should do nothing, i.e. trivial
maybe the condition is a bit wrong?
| and residual_output_mode == ScatterMode.TP_ATTN_FULL | ||
| ): | ||
| return CommunicateSimpleFn._scattered_to_tp_attn_full | ||
| return CommunicateSimpleFn._gather_hidden_states_and_residual |
There was a problem hiding this comment.
the if looks like we only gathere residual, so wondering maybe if condition is wrong (or the function name is wrong)
fzyzcjy
approved these changes
May 27, 2025
|
@ch-wan maybe you should run "pre-commit run --all-files" to pass the lint tests |
zhyncs
approved these changes
Jun 16, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
Motivation
Fix one issue when we use DP for dense FFNs and EP (not DeepEP) for sparse FFNs. Related issue: #6297
Modifications
Checklist