MoE expert parallelism + sequence parallelism#45408
Merged
3outeille merged 10 commits intorefactor-tp-dtensorfrom Apr 14, 2026
Merged
MoE expert parallelism + sequence parallelism#454083outeille merged 10 commits intorefactor-tp-dtensorfrom
3outeille merged 10 commits intorefactor-tp-dtensorfrom
Conversation
- Add PackedColwiseParallel for fused gate_up_proj weights - Add MoEExpertsParallel with per-expert DTensor sharding - Add PrepareModuleInputOutput for SP allgather/split hooks - Add _AllReduceBackward for MoE routing weight gradients - Extend TPStyle with moe_experts, packed_colwise, activation, module kinds - _StridedShard handling in core_model_loading for interleaved weights - MoE model configs: mixtral, deepseek_v3, qwen3 with SP plans - DTensor rotary_pos_emb guard for mixtral
3outeille
force-pushed
the
refactor-tp-dtensor
branch
from
34a5085 to
eb428cc
Compare
3outeille
force-pushed
the
moe-sequence-parallel
branch
from
e04c7d9 to
24ca327
Compare
3outeille
force-pushed
the
refactor-tp-dtensor
branch
from
eb428cc to
e0c4e06
Compare
3outeille
force-pushed
the
moe-sequence-parallel
branch
from
24ca327 to
7f297e0
Compare
# Conflicts: # src/transformers/integrations/tensor_parallel.py
# Conflicts: # src/transformers/integrations/tensor_parallel.py
The _IdentityOp class (added by PR #44983) was accidentally deleted during the MoE expert parallelism work. It is needed by finegrained_fp8.py and metal_quantization.py as a pass-through reverse_op for dequantize operations. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3outeille
force-pushed
the
moe-sequence-parallel
branch
from
5031188 to
01866b8
Compare
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: deepseek_v3, dots1, mixtral, nanochat, qwen3, qwen3_5, qwen3_5_moe, qwen3_moe, qwen3_next, qwen3_omni_moe, qwen3_vl, qwen3_vl_moe, youtu |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
* from_pretrained orchestration + save/load - Add gather_full_state_dict() for DTensor→full tensor saving - Add convert_strided_to_shard() / restore_strided_from_shard() for DCP - Add _redistribute_dtensor() helper - Full distributed_config integration in from_pretrained/save_pretrained - Rename apply_fsdp2 → apply_fully_shard_data_parallel - save_optimizer() / load_optimizer() in distributed/utils - Trainer integration with distributed_config - Updated FSDP and TP tests for new orchestration API - DTensor shard-on-read test updates * revert distributed utils * eaaea * all tests for core modeling are passing * populate import from init for tp * ruff * ruff
Contributor
|
View the CircleCI Test Summary for this PR: https://huggingface.co/spaces/transformers-community/circle-ci-viz?pr=45408&sha=bbf3ab |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Summary
PackedColwiseParallel,MoEExpertsParallel,PrepareModuleInputOutput,_AllReduceBackwardcustom ParallelStyle subclassesTPStylewithmoe_experts,packed_colwise,activation,module,loss_parallelkinds_StridedShardhandling incore_model_loading.pyfor interleavedgate_up_projweightsmixtral,deepseek_v3,qwen3with sequence parallelism plansPart of the distributed training API chain: #44989
Chain:
main ← #44989 ← #44083 ← #44974 ← #45028 ← this PR ← orchestration+save PRReview question
Are the custom
ParallelStylesubclasses correct for expert sharding + sequence parallelism?Test plan