Skip to content

Pull requests: pytorch/torchtitan

New pull request New
Clear current search query, filters, and sorts
28 Open 36 Closed
28 Open 36 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

fix(MoE): Apply grad_placements=Partial() to router to_local for TP CLA Signed This label is managed by the Meta Open Source bot.
#2388 by fatih-uzlmz was closed Feb 21, 2026 Loading… updated Feb 21, 2026
1
14
Add run-to-run determinism testing to H100 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#2339 by xmfan was closed Feb 17, 2026 Loading… updated Feb 17, 2026
12
initial skeleton CLA Signed This label is managed by the Meta Open Source bot.
#2376 by daniellepintz was closed Feb 17, 2026 • Draft updated Feb 17, 2026
39
Fix for incompatible ReduceOp.PREMUL_SUM on XPU devices CLA Signed This label is managed by the Meta Open Source bot.
#2332 by saforem2 was closed Feb 6, 2026 Loading… updated Feb 11, 2026
10
fix imports in components/checkpoint.py CLA Signed This label is managed by the Meta Open Source bot.
#1844 by saforem2 was closed Feb 9, 2026 Loading… updated Feb 9, 2026
4
feat(moe): add topk_before_score routing and use_router_bias support CLA Signed This label is managed by the Meta Open Source bot.
#2212 by eous was closed Feb 5, 2026 Loading… updated Feb 5, 2026
@shuhuayu
28
fixed validation error when using flash attention CLA Signed This label is managed by the Meta Open Source bot.
#2142 by francesco-bertolotti was closed Jan 27, 2026 Loading… updated Jan 27, 2026
1
13
feat(moe): add use_expert_bias config for optional expert biases CLA Signed This label is managed by the Meta Open Source bot.
#2214 by eous was closed Jan 20, 2026 Loading… updated Jan 20, 2026
4
[Float8] Fix intermediate Float8 checkpoint loading CLA Signed This label is managed by the Meta Open Source bot.
#1414 by jquesnelle was closed Jan 17, 2026 Loading… updated Jan 17, 2026
6
refactor: small improvement in lr_scheduler.py
#2147 by nabil-devs was closed Dec 18, 2025 Loading… updated Dec 18, 2025
2
Fix torch.compile recompilation issue with HF modeling + TP CLA Signed This label is managed by the Meta Open Source bot.
#2130 by 3outeille was closed Dec 15, 2025 Loading… updated Dec 15, 2025
8
Warn that SAC + Compile for MoE models is not yet supported CLA Signed This label is managed by the Meta Open Source bot.
#2052 by xmfan was closed Dec 5, 2025 Loading… updated Dec 5, 2025
10
Workaround AC HOP mutation issue when tracing token dispatch CLA Signed This label is managed by the Meta Open Source bot.
#1984 by xmfan was closed Dec 5, 2025 • Draft updated Dec 5, 2025
1
5
[precompile] add ability to precompile torchtitan models CLA Signed This label is managed by the Meta Open Source bot.
#2092 by bobrenjc93 was closed Dec 2, 2025 Loading… updated Dec 2, 2025
8
[Compiler Toolkit] Integration test CLA Signed This label is managed by the Meta Open Source bot.
#1953 by SherlockNoMad was closed Oct 30, 2025 Loading… updated Oct 30, 2025
2
Adding prefetching of first shards to train script when fsdp enabled CLA Signed This label is managed by the Meta Open Source bot.
#1955 by chelsea0x3b was closed Oct 29, 2025 Loading… updated Oct 29, 2025
2
Annotate dsv3 with layer_id CLA Signed This label is managed by the Meta Open Source bot.
#1908 by SherlockNoMad was closed Oct 27, 2025 Loading… updated Oct 27, 2025
2
Update vocab_size for debugmodel_moe of qwen3 moe CLA Signed This label is managed by the Meta Open Source bot.
#1864 by wmhst7 was closed Oct 14, 2025 Loading… updated Oct 14, 2025
5
handle unable to load ft checkpoint CLA Signed This label is managed by the Meta Open Source bot.
#1729 by tushar00jain was closed Sep 25, 2025 Loading… updated Sep 25, 2025
2
temporarily removed cudnn attention backend CLA Signed This label is managed by the Meta Open Source bot.
#1717 by danielvegamyhre was closed Sep 18, 2025 Loading… updated Sep 18, 2025
13
Adding small model configurations to torchtitan CLA Signed This label is managed by the Meta Open Source bot. fb-exported meta-exported
#1719 by Shagun-G was closed Sep 18, 2025 Loading… updated Sep 18, 2025
4
[Feat] Gradient sync is turned off during gradient accumulation. CLA Signed This label is managed by the Meta Open Source bot.
#1710 by EquationWalker was closed Sep 15, 2025 Loading… updated Sep 15, 2025
4
remove dead code CLA Signed This label is managed by the Meta Open Source bot.
#1450 by tushar00jain was closed Jul 31, 2025 Loading… updated Jul 31, 2025
5
[simple_fsdp] apply bucketing ag/rs passes, reordering collectives, sink CLA Signed This label is managed by the Meta Open Source bot.
#1464 by IvanKobzarev was closed Jul 25, 2025 Loading… updated Jul 25, 2025
10
[DSV3] Explicitly convert to bfloat16 when use grouped mm CLA Signed This label is managed by the Meta Open Source bot.
#1367 by wwwjn was closed Jul 8, 2025 Loading… updated Jul 19, 2025
6
ProTip! no:milestone will show everything without a milestone.