Skip to content

Add distributed training CI#42765

Draft
3outeille wants to merge 17 commits intomainfrom
v5-distributed-training-ci
Draft

Add distributed training CI#42765
3outeille wants to merge 17 commits intomainfrom
v5-distributed-training-ci

Conversation

@3outeille
Copy link
Copy Markdown
Member

@3outeille 3outeille commented Dec 10, 2025

it needs several PR:

  • Add FSDP v2 (with dtensor for now) natively to Transformers (not rely on transformer anymore) => needs to compare correctness through DDP
  • Check save/load of TP/FSDP/FSDP+TP
  • distributed training ci
    • use gloo backend
    • no subprocess
    • check if it bitwise convergent in terms of loss & grad_norm
    • need to add FSDP torchtitan + TP from transformers
    • Fix related bug

@HuggingFaceDocBuilderDev
Copy link
Copy Markdown

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.

@github-actions
Copy link
Copy Markdown
Contributor

View the CircleCI Test Summary for this PR:

https://huggingface.co/spaces/transformers-community/circle-ci-viz?pr=42765&sha=7b744c

3outeille and others added 4 commits February 2, 2026 16:15
- Introduced `TrainingConfigMixin` to share hyperparameters between `TrainingTesterMixin` and `TrainingDistributedTesterMixin`.
- Updated `TrainingDistributedTesterMixin` to inherit from `TrainingConfigMixin` and adjusted training parameters for faster distributed tests.
- Enhanced documentation for clarity on the purpose of each mixin.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants