[Feat] [FSDP2] [SFT] [ExpertParallel] [DeepSeekV4] fsdp2 dpsk v4 flash 0731 support on only 1Node with 8*H200 - #10197
Open
mugglewei97 wants to merge 2 commits into
Open
mugglewei97 wants to merge 2 commits into
mugglewei97 wants to merge 2 commits into
Conversation
added 2 commits
September 16, 2026 19:29
hjh0119
added
the
postpone
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
PR type
New Feature
More Models or Datasets Support
PR information
为混合专家（MoE）模型增加了基于 FSDP2 后端、具备普适性的专家并行EP
使得仅在单节点8*H200上即可实现full sft训练
由于MoE模型绝大部分权重在experts，而experts的激活是十分稀疏的，为避免fsdp2在fully shard下，
每次forward/pre-fetch时all gather整个decoder layer的完整权重，通过ignore modules将experts部分从fsdp2中解耦，为其实A2A通信和local expert计算
另，通过令device=meta，解决了大尺寸模型初始化加载权重时的oom问题
目前在8*H200即可加载DeepSeek-V4-Flash-0731
Experiment results
Training script which has been actual tested is added
Paste your experiment result here(if needed).