Skip to content

Pull requests: pytorch/torchtitan

New pull request New
304 Open 2,160 Closed
304 Open 2,160 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

[GraphTrainer] Remove SAC peak-memory test from H100 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3170 opened Apr 30, 2026 by SherlockNoMad Contributor Loading…
1 task
[MoE][4/n] deprecate expert tensor parallel (ETP) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3167 opened Apr 29, 2026 by acisseJZhong Contributor Loading…
1 task
9
[cpu-offloading] Implement prefetching via env var configs ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3166 opened Apr 29, 2026 by mlazos Contributor Loading…
8
[Module][Qwen3-VL] Full config-based sharding + full_dtensor ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3165 opened Apr 29, 2026 by fegin Contributor • Draft
[graph_trainer] Generalize minimal_fx_tracer to module + optimizer roots ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3164 opened Apr 29, 2026 by tugsbayasgalan Contributor Loading…
Temporarily disable torchcomms TP+PP+compile test ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3163 opened Apr 29, 2026 by mori360 Contributor Loading…
[graph_trainer] Re-enable ChunkedCELoss via compute_traceable_grads ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3162 opened Apr 29, 2026 by tugsbayasgalan Contributor Loading…
3
[NOT READY FOR REVIEW][Full DTensor] Enable full_dtensor for MoE models (qwen3, llama4, gpt_oss, deepseek_v3) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3161 opened Apr 29, 2026 by fegin Contributor • Draft
[NOT READY FOR REVIEW][Module][MoE] Config-based MoE sharding (qwen3, llama4, gpt_oss, deepseek_v3) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3160 opened Apr 29, 2026 by fegin Contributor • Draft
[NOT READY FOR REVIEW][Full DTensor] Config-based Full DTensor for Llama3 ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3159 opened Apr 29, 2026 by fegin Contributor • Draft
24
[CI] Move lychee link checker to nightly workflow ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3158 opened Apr 29, 2026 by fegin Contributor Loading…
10
[graph_trainer] FSDP AG RS overlap ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3156 opened Apr 29, 2026 by yiming0416 Contributor Loading…
6
[will close, do not review] Revert "[Module] Remove LocalMapInnerAttention" and "[rl] Removed pat… ci-no-td ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3154 opened Apr 29, 2026 by daniellepintz Contributor Loading…
[will close, do not review] test rl ci ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3153 opened Apr 29, 2026 by daniellepintz Contributor Loading…
refactor get_attention_masks to use positions instead of eos_id ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3149 opened Apr 29, 2026 by liangel-02 Contributor Loading…
2
11
[MoE][SAC] Use deterministic ops in MoE routing ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3146 opened Apr 28, 2026 by songhappy Loading…
3
Improve compilation time (reduce from ~50 seconds to ~15s for vLLM) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3145 opened Apr 28, 2026 by Lucaskabela Contributor • Draft
2
27
[CI] Run integration tests in parallel ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3144 opened Apr 28, 2026 by yuankaichen-amd Loading…
6
[WIP] Enable DP-to-EP for MoE inference ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3143 opened Apr 28, 2026 by wwwjn Contributor Loading…
1
Enable TP2EP for MoE inference in vLLM wrapper ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3142 opened Apr 28, 2026 by wwwjn Contributor Loading…
19
[torchtitan] Fix deepseek_v3_16b config to use deepep comm backend ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3139 opened Apr 28, 2026 by anshul-si Contributor Loading…
3
[DO NOT MERGE] Update README.md ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3136 opened Apr 28, 2026 by jithunnair-amd Contributor • Draft
1
[GraphTrainer][AutoDev] Fuse RoPE kernels via regional Inductor compilation ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3134 opened Apr 28, 2026 by SherlockNoMad Contributor • Draft
2 of 4 tasks
1
[GraphTrainer][AutoDev] Fuse RMSNorm kernels via regional Inductor compilation ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3132 opened Apr 28, 2026 by SherlockNoMad Contributor Loading…
3 tasks done
8
Fix commands in DSV3 readme ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3116 opened Apr 27, 2026 by danielvegamyhre Contributor Loading…
ProTip! Mix and match filters to narrow down what you’re looking for.