Skip to content

Pull requests: pytorch/torchtitan

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
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

[Qwen3.5] Fix MoE expert weight layouts in state dict adapter CLA Signed This label is managed by the Meta Open Source bot.
#4373 opened Aug 29, 2026 by taking-lying-flat Contributor Loading…
[DO NOT REVIEW] Add MXFP8 recipes for the qwen3 and gpt_oss MoE models CLA Signed This label is managed by the Meta Open Source bot.
#4372 opened Aug 28, 2026 by anijain2305 Contributor 5/5 Draft
[DO NOT REVIEW] Handle unevenly sharded expert weights in FSDP compute weights CLA Signed This label is managed by the Meta Open Source bot.
#4371 opened Aug 28, 2026 by anijain2305 Contributor 4/5 Draft
Pass the expert weight, not its transpose, to the grouped-mm seam CLA Signed This label is managed by the Meta Open Source bot.
#4368 opened Aug 28, 2026 by anijain2305 Contributor 2/5 Loading…
[flux] Count both attention matmuls in the FLOPs estimate CLA Signed This label is managed by the Meta Open Source bot.
#4366 opened Aug 28, 2026 by alanhuangyoo Contributor Loading…
[rl] RendererConfig silently drops the removed preserve_* renderer knobs ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4365 opened Aug 28, 2026 by andylizf Loading…
switch GDN to use attn-gym kernels for titanRL ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4363 opened Aug 28, 2026 by liangel-02 Contributor Loading…
[ROCm][CI] Re-enable ROCm CI & make CI changes as per new CI design ciflow/rocm CLA Signed This label is managed by the Meta Open Source bot. module: rocm
#4361 opened Aug 27, 2026 by akashveramd Collaborator Draft
Enable CudaGraph Annotations CLA Signed This label is managed by the Meta Open Source bot.
#4359 opened Aug 27, 2026 by SherlockNoMad Contributor Loading…
[rl] Add optional Verifiers rollout integration ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4356 opened Aug 27, 2026 by wwwjn Contributor Draft
[models] Upgrade Qwen3.5 implementation to Qwen3.8 ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4355 opened Aug 27, 2026 by wwwjn Contributor Loading…
Removed HF trainer CLA Signed This label is managed by the Meta Open Source bot.
#4354 opened Aug 27, 2026 by jinsooihm Contributor Loading…
[DistMuon] Support tensor-parallel storage layouts CLA Signed This label is managed by the Meta Open Source bot.
#4353 opened Aug 27, 2026 by shuhuayu Contributor Draft
[WIP] kda in torchtitan (vllm integration) ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4352 opened Aug 27, 2026 by liangel-02 Contributor Loading…
[kimi k3] Support sample packing in KDA CLA Signed This label is managed by the Meta Open Source bot.
#4347 opened Aug 27, 2026 by JavaZeroo Contributor Draft
1 of 3 tasks
[DO NOT REVIEW] Add FSDP-managed MXFP8 grouped-expert training CLA Signed This label is managed by the Meta Open Source bot.
#4344 opened Aug 27, 2026 by anijain2305 Contributor 3/5 Draft
[SDC] Integrate deterministic replay into Trainer CLA Signed This label is managed by the Meta Open Source bot.
#4334 opened Aug 26, 2026 by xmfan Member Loading…
[SDC] Add deterministic replay engine CLA Signed This label is managed by the Meta Open Source bot.
#4333 opened Aug 26, 2026 by xmfan Member Loading…
Fix rowwise TP bias gradients CLA Signed This label is managed by the Meta Open Source bot.
#4332 opened Aug 26, 2026 by taking-lying-flat Contributor Loading…
refactor max_context_length ciflow/rl CLA Signed This label is managed by the Meta Open Source bot.
#4328 opened Aug 26, 2026 by francesco-bertolotti Contributor Loading…
Pass --test_scope=real_pg_required in the PyTorch CI model tests CLA Signed This label is managed by the Meta Open Source bot.
#4327 opened Aug 26, 2026 by weifengpy Contributor Draft
GraphTrainer spmd_types backend CLA Signed This label is managed by the Meta Open Source bot.
#4325 opened Aug 25, 2026 by pianpwk Contributor Loading…
[WIP][CP] Move collectives into attention kernels CLA Signed This label is managed by the Meta Open Source bot.
#4322 opened Aug 25, 2026 by fegin Contributor Loading…
[spmd_types] Save logical shapes for padded parameters CLA Signed This label is managed by the Meta Open Source bot.
#4320 opened Aug 25, 2026 by pianpwk Contributor Draft
[spmd_types] Support padded uneven vocabulary logits CLA Signed This label is managed by the Meta Open Source bot.
#4319 opened Aug 25, 2026 by pianpwk Contributor Draft
ProTip! no:milestone will show everything without a milestone.