Skip to content

Pull requests: pytorch/torchtitan

New pull request New
353 Open 2,528 Closed
353 Open 2,528 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

[rl] Fix batch invariant logprob calculation by forcing vllm use trainer's function ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3629 opened Jun 11, 2026 by wwwjn Contributor Loading…
Memory snapshot: python-only stacks for fast dumps + configurable max_entries ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3628 opened Jun 11, 2026 by SherlockNoMad Contributor Loading…
[MoE] Use CPU split-size sum for EP permute output size ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3627 opened Jun 10, 2026 by sanketpurandare Contributor Loading…
12
Add StickySessionRoutingStrategy for GeneratorRouter ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3625 opened Jun 10, 2026 by pzhan9 Contributor Loading…
7
Add num_generators field to RLTrainer.Config ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3624 opened Jun 10, 2026 by pzhan9 Contributor Loading…
2
[Checkpointer] Remove the dependencies on PyTorch distributed state_dict APIs ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3623 opened Jun 10, 2026 by fegin Contributor Loading…
3
[graph_trainer] Device-hoist CooR artifacts (drop --virtual-local-rank) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3620 opened Jun 10, 2026 by bobrenjc93 Contributor • Draft
4 tasks
[graph_trainer] dsv3 run: local_batch_size=16 + Run 6 batch-scaling results CLA Signed This label is managed by the Meta Open Source bot.
#3618 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Match Eager FSDP bucket order ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3617 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Force cudagraph for MinimalAsyncEP on H100 + Run 4 results ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3616 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
Add MinimalAsyncEP + offset aware swiglu kernels (#3561) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3615 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Enable ChunkedCELoss for deepseek_v3 and qwen3 (#3248) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3614 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Replace stable_topological_sort with _move_overlap_nodes in fsdp_passes ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3613 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Add DSv3 scaling run/profiling helper script CLA Signed This label is managed by the Meta Open Source bot.
#3612 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Match Eager FSDP bucket order ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3611 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Force cudagraph for MinimalAsyncEP on H100 + Run 4 results ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3610 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
Add MinimalAsyncEP + offset aware swiglu kernels (#3561) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3609 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Enable ChunkedCELoss for deepseek_v3 and qwen3 (#3248) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3608 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Replace stable_topological_sort with _move_overlap_nodes in fsdp_passes ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3607 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
[graph_trainer] Add DSv3 scaling run/profiling helper script CLA Signed This label is managed by the Meta Open Source bot.
#3606 opened Jun 10, 2026 by SherlockNoMad Contributor • Draft
1
[GraphTrainer] Fix test_deterministic: update model hash and calling update_from_config ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3605 opened Jun 10, 2026 by wwwjn Contributor Loading…
2
[Bug] Fix MoE SP token combine indices bug Something isn't working ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot. high priority
#3604 opened Jun 10, 2026 by Matrix-Z97 Loading…
@wwwjn
4
[flex_shard] Add DeepSeek V3 eager training entry point ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3603 opened Jun 10, 2026 by weifengpy Contributor • Draft
[rl] Search-R1: multi-turn retrieval-augmented GRPO ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3602 opened Jun 10, 2026 by yichuan-w Member Loading…
3
Add deterministic topk for MoE routing ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3600 opened Jun 10, 2026 by sanketpurandare Contributor Loading…
3
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.