[SPMD] use UNKNOWN as default sharded data sharding type only for auto-sharding #6846
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
yeounoh
added
the
DO_NOT_MERGE_YET
yeounoh
changed the title
yeounoh
force-pushed
the
spmd_inference_perf_regression
branch
3 times, most recently
from
938ced4 to
251d285
Compare
|
I can confirm that with this PR, the LLaMa 2 inference numbers are back to pre-regression levels, cc @yeounoh |
JackCaoG
reviewed
Mar 28, 2024
Comment on lines
+252
to
+255
| ShardingSpecPtr current_sharding = sharding_spec(); | ||
| if (!current_sharding || overwrite || | ||
| current_sharding->sharding.type() == xla::OpSharding::REPLICATED || | ||
| current_sharding->sharding.type() == xla::OpSharding::UNKNOWN) { |
There was a problem hiding this comment.
this is just a clean up right? I don't see any functionality change but want to double check.
JackCaoG
approved these changes
Mar 28, 2024
yeounoh
force-pushed
the
spmd_inference_perf_regression
branch
from
251d285 to
da66da3
Compare
yeounoh
removed
the
DO_NOT_MERGE_YET
yeounoh
added a commit
that referenced
this pull request
Mar 29, 2024
This was referenced Mar 29, 2024
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Outdated suggestions cannot be applied.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.
Using UNKNOWN as default sharded data sharding type degraded SPMD inference performance. I added a comment to do a follow-up investigation, to see if we can completely move to Unknown for non-auto-sharding route as well.
Test plan: