[JAX] Rework amax reduction over TPSP #2218
Open
+35
−25
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
Signed-off-by: Phuong Nguyen <[email protected]>
Signed-off-by: Phuong Nguyen <[email protected]>
jberchtold-nvidia
approved these changes
Sep 30, 2025
| sequence_dim = 0 if batch_sequence_transpose else 1 | ||
| # Run AR across TPSP only when tensor-sequence is detected in the input spec | ||
| if amax_scope is AmaxScope.TPSP and x_spec[sequence_dim] == gmesh.tpsp_resource: | ||
| amax = lax_paral_op(amax, jax.lax.pmax, gmesh.tpsp_resource, mesh) |
There was a problem hiding this comment.
Previously, we would also AR in the gmesh.tp_resource dim. To keep the previous behavior, do we still need to reduce on TP if amax_scope is TPSP but the sequence dim is TP not TPSP? Or is it okay to only reduce when tpsp is active?
There was a problem hiding this comment.
We only do AG the sequence dimension before the GEMM in TPSP, thus amax reduction is needed.
If users use TP for TPSP, there are existing warnings in the GEMM op to warn users to switch to TPSP.
|
/te-ci JAX L1 |
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Outdated suggestions cannot be applied.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.
Description
In this PR, we add an additional check to detect whether the input is partitioned in the sequence dimension and only trigger the amax reduction across TPSP in that case.
With this change, the
using_global_amax_of_xis no longer needed.Type of change
Checklist: