Skip to content

Pull requests: pytorch/torchtitan

New pull request New
307 Open 2,358 Closed
307 Open 2,358 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

[graph_trainer] Fix H100 CI failure from DeepEP compilation break ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3390 opened May 17, 2026 by SherlockNoMad Contributor Loading…
2 tasks
2
[MoE][6/n] Extract local_reorder, split DeepEP/HybridEP dispatchers ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3389 opened May 17, 2026 by acisseJZhong Contributor Loading…
3 tasks
6
Add xpugraph pass for graph trainer ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3388 opened May 17, 2026 by jemitche1 • Draft
1
[MoE][5/n] Refactor MoE to clean DTensor boundaries for shared/routed experts ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3386 opened May 17, 2026 by acisseJZhong Contributor Loading…
4 tasks done
24
Update ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3385 opened May 17, 2026 by acisseJZhong Contributor Loading…
Fix memory snapshot pickle protocol for compatibility with memory visualizers ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3375 opened May 16, 2026 by francesco-bertolotti Contributor Loading…
4
Re-enable compile for gpt-oss integration tests ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3373 opened May 15, 2026 by aditvenk Contributor Loading…
[qwen3_5] evolve qwen3_vl to qwen3_5 ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3371 opened May 15, 2026 by shuhuayu Contributor Loading…
20
[graph_trainer] Use separate EP process groups for overlap ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3369 opened May 15, 2026 by sanketpurandare Contributor Loading…
4
[notforland] bitwise infra ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3366 opened May 15, 2026 by pianpwk Contributor • Draft
[rl] fix CI timeout issue by properly tear down for vllm engine V2 ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3365 opened May 15, 2026 by wwwjn Contributor Loading…
2
to_local() for ChunkedCELoss + no LP ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3364 opened May 15, 2026 by pianpwk Contributor Loading…
8
[graph_trainer] Add EP overlap eager chunking scaffolding ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3363 opened May 15, 2026 by sanketpurandare Contributor Loading…
[graph_trainer] Support hinted symbolic input dims in tracing ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3362 opened May 15, 2026 by sanketpurandare Contributor Loading…
4
[graph_trainer] Add DeepSeek V3 16B SDPA config ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3361 opened May 15, 2026 by sanketpurandare Contributor Loading…
Enables HybridEP torch.compile support and adds eager integration tests ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3360 opened May 14, 2026 by syed-ahmed Collaborator Loading…
7
[rl] Route RL CI with pytest markers ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3357 opened May 14, 2026 by felipemello1 Contributor Loading…
25
Fix optimizer state and module state coupling ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3356 opened May 14, 2026 by tugsbayasgalan Contributor Loading…
1
[graph_trainer] Defer cudagraph compatibility check to pass execution… ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3355 opened May 14, 2026 by IvanKobzarev Contributor Loading…
[graph_trainer] Add bucketing ops to precompile serialization filter ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3354 opened May 14, 2026 by IvanKobzarev Contributor Loading…
[graph_trainer] Make JointManualOverlapScheduler standalone with direct node moves ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3352 opened May 14, 2026 by SherlockNoMad Contributor • Draft
1 of 3 tasks
Scope FlexShard recompute policy by provenance ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3349 opened May 14, 2026 by weifengpy Contributor • Draft
Support grouped FlexShard reshard-after-forward buckets ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3348 opened May 14, 2026 by weifengpy Contributor • Draft
[rl] Add Batcher in RL Loop ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3347 opened May 14, 2026 by wwwjn Contributor Loading…
48
[graph_trainer] Rework full_inductor_compilation_pass via regional_inductor + CPU attr migration ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3346 opened May 13, 2026 by tugsbayasgalan Contributor Loading…
ProTip! Filter pull requests by the default branch with base:main.