Better sharding for dsv3 moe layer #2373
Merged
+53
−14
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
suexu1025
requested review from
RissyRan,
michelle-yooh,
gagika and
richjames0
as code owners
suexu1025
changed the title
gobbleturk
reviewed
Sep 19, 2025
RissyRan
reviewed
Sep 19, 2025
|
suexu1025
requested review from
khatwanimohit,
bvandermoon,
vipannalla,
shralex,
yangyuwei,
SurbhiJainUSC,
hengtaoguo,
A9isha,
aireenmei,
NuojCheng and
jiangjy1982
as code owners
gobbleturk
reviewed
Sep 22, 2025
gobbleturk
reviewed
Sep 22, 2025
gobbleturk
approved these changes
Sep 23, 2025
Thanks! I meant to say the original links, so that we could directly to see profiles or anything. One example here, usually put original links for reviewers. |
RissyRan
reviewed
Sep 23, 2025
RissyRan
reviewed
Sep 23, 2025
update refactor update Add optional config Explicitly shard input tensors across mesh devices Run on 0.7.2 candidate image Fix typo in image tag Revert to use latest tag update test for new jax version Remove sharding rules for q_lora and kv_lora from base.yml update with configs clean up update
suexu1025
force-pushed
the
qinwen/update_sharding_moe
branch
from
b0cbe7f to
859abdf
Compare
RissyRan
reviewed
Sep 24, 2025
| if raw_keys["fsdp_shard_on_exp"] and raw_keys["num_experts"] % raw_keys["ici_fsdp_parallelism"]!=0: | ||
| raise ValueError("fsdp_shard_on_exp requires num_experts is divisiable by ici_fsdp_parallelism.") | ||
| if raw_keys["fsdp_shard_on_exp"] and (using_tensor_parallelism(raw_keys) or useing_expert_parallelism(raw_keys)): | ||
| raise ValueError("fsdp_shard_on_exp requires ici_expert_parallelism = 1 and ici_tensor_parallelism/ici_tensor_transpose_parallelism = 1.") |
There was a problem hiding this comment.
Nit: probably just say fsdp_shard_on_exp does not support EP and TP shardings?
RissyRan
reviewed
Sep 24, 2025
| axis=-1, | ||
| kernel_init=self.kernel_init, | ||
| kernel_axes=("embed", "q_lora"), | ||
| kernel_axes=("embed", "q_lora_up_proj"), |
There was a problem hiding this comment.
This part is helpful for all MLA models or only DS V3?
cc @richjames0 another case.
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
land sharding strategy for moe layer.
--fsdp_shard_on_exp=trueto enable shard fsdp on num_expert dim[43s]
If the change fixes a bug or a Github issue, please include a link, e.g.,:
FIXES: b/123456
FIXES: #123456
Notice 1: Once all tests pass, the "pull ready" label will automatically be assigned.
This label is used for administrative purposes. Please do not add it manually.
Notice 2: For external contributions, our settings currently require an approval from a MaxText maintainer to trigger CI tests.
Tests
Please describe how you tested this change, and include any instructions and/or
commands to reproduce.
Checklist
Before submitting this PR, please make sure (put X in square brackets):