Skip to content

Pull requests: pytorch/torchtitan

New pull request New
251 Open 1,862 Closed
251 Open 1,862 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

[MoE Rewrite 1/n] Use local map for torch.histc and torch.gather, and use DTensor for router ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2770 opened Mar 31, 2026 by acisseJZhong • Draft
4
[WIP] Sft multiturn CLA Signed This label is managed by the Meta Open Source bot.
#2769 opened Mar 31, 2026 by haydn-jones • Draft
1
[graph_trainer] Add remat pass and torch.no_grad() execution to minimal_fx_tracer ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2766 opened Mar 31, 2026 by tugsbayasgalan Loading…
7
[WIP][rl] test rocm CI ciflow/rocm-mi300 ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot. module: rocm
#2765 opened Mar 31, 2026 by wwwjn Loading…
3
[graph_trainer] Add cudagraph support for Inductor-compiled and non-compiled graphs ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2762 opened Mar 31, 2026 by bobrenjc93 • Draft
[graph_trainer] Annotate backward nodes with remat_pass_tag for AC rematerialization ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2760 opened Mar 31, 2026 by tugsbayasgalan Loading…
[graph_trainer] Replace trace_module/run_traced_module with aot_function API ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2759 opened Mar 31, 2026 by tugsbayasgalan Loading…
[compile] Share SimpleFSDP wrapper class across same-type module instances CLA Signed This label is managed by the Meta Open Source bot.
#2754 opened Mar 30, 2026 by anijain2305 Loading…
5
[graph_trainer] Replace trace_module/run_traced_module with aot_function API ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2753 opened Mar 30, 2026 by tugsbayasgalan Loading…
29
[rl] Remove ref_model from PolicyTrainer, make KL penalty optional ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2750 opened Mar 30, 2026 by daniellepintz Loading…
1
[graph_trainer] aot_nested_region: add kwargs support ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2746 opened Mar 29, 2026 by tugsbayasgalan Loading…
[graph_trainer] Add aot_nested_region for invoke_subgraph deduplication ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2745 opened Mar 29, 2026 by tugsbayasgalan Loading…
[graph_trainer] Extract shared model setup helpers into torchtitan/model_setup.py ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2743 opened Mar 28, 2026 by bobrenjc93 • Draft
gap for per-layer compiler (with moe) ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot. high priority
#2741 opened Mar 27, 2026 by weifengpy Loading…
3
Add shard for FSDP and HSDP in features test workflow. ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2740 opened Mar 27, 2026 by akashveramd • Draft
@akashveramd
[graph_trainer] Differentiate through original subgraph_fn in backward ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2734 opened Mar 27, 2026 by tugsbayasgalan Loading…
Add bf16 optimizer state support via step pre-hook CLA Signed This label is managed by the Meta Open Source bot.
#2732 opened Mar 27, 2026 by vivekgoe • Draft
13
[graph_trainer] aot_nested_region: add kwargs support ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2731 opened Mar 27, 2026 by tugsbayasgalan Loading…
[graph_trainer] Add aot_nested_region for invoke_subgraph deduplication ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2730 opened Mar 27, 2026 by tugsbayasgalan Loading…
Add async logging ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2728 opened Mar 27, 2026 by drisspg Loading…
2
Enable graph PP for local_map_deepseek_v3 ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2727 opened Mar 27, 2026 by sanketpurandare • Draft
Add graph PP infrastructure for autoparallel ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2726 opened Mar 27, 2026 by sanketpurandare • Draft
Fix DeepSeekV3Model for Configurable build pattern ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2725 opened Mar 27, 2026 by sanketpurandare • Draft
1
Refactor pipeline_parallel.py for graph PP reuse ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2724 opened Mar 27, 2026 by sanketpurandare • Draft
5
[Perf] Fusing router f32 bmm into a triton kernel CLA Signed This label is managed by the Meta Open Source bot. kernel optimization
#2717 opened Mar 26, 2026 by chelsea0x3b Loading…
ProTip! Type g i on any issue or pull request to go back to the issue listing page.