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Outdated suggestions cannot be applied.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.
What this PR does / why we need it?
After #1229 , multistream moe with both mc2 and TP is broken, due to
all_gatheroperation in moe layer used to communicate the summation ofrouter_hidden_statesandshared_hidden_states, but onlyrouter_hidden_statesnow, causing a shape mismatch.Fix this by adding yet another
all_gatheroperation forshared_hidden_states.By the way, all these
all_gatheroperation exists due to a joint effect of:all_reduceinstead ofall_gather+reduce_scatterfor TP communication and,splitoperations at the entry of moe layer.This problem will be addressed by #1194 -like optimizations, after that is merged, both
splitandall_gatherhere can be removed.Does this PR introduce any user-facing change?
No.
How was this patch tested?
Tested both offline, and by graph mode mla e2e testcase.