All issues

Search results

TorchTitan model support for TorchComms backend

wontfixThis will not be worked onThis will not be worked on

Status: Open.

#4839 In pytorch/torchtitan;

· newtdms opened on Sep 22, 2026

[CI] tests/unit_tests/rl is not collected by CPU/GPU unit jobs

Status: Open.

#4828 In pytorch/torchtitan;

· YeonwooSung opened on Sep 22, 2026

Restore the training loss for the DeepSeek V4 indexer

Status: Open.

#4802 In pytorch/torchtitan;

· HaoranYu111 opened on Sep 20, 2026

DeepSeek V4 ignores packed-document boundaries in sparse attention

Status: Open.

#4801 In pytorch/torchtitan;

· HaoranYu111 opened on Sep 20, 2026

[transformers_modeling_backend]Regression of #3775: rotary inv_freq is left uninitialized after to_empty()

Status: Open.

#4783 In pytorch/torchtitan;

· Jiarwang77 opened on Sep 18, 2026

[Bug] An asynchronous TorchFT healing export reshards an in-flight FSDP2 parameter

Status: Open.

#4753 In pytorch/torchtitan;

· jojoinfra opened on Sep 16, 2026

[RL] #4535 (fused gate-up default) makes RL weight sync copy the whole FFN projection every step

Status: Open.

#4706 In pytorch/torchtitan;

· AlirezaShamsoshoara opened on Sep 15, 2026

[RFC][DistMuon] Handle fused weights correctly for Kimi K3 per-head Muon

Status: Open.

#4692 In pytorch/torchtitan;

· JavaZeroo opened on Sep 15, 2026

[graph_trainer] SAR omits in-place writes to recomputed buffers, producing incorrect gradients

Status: Open.

#4688 In pytorch/torchtitan;

· johnjunjun7 opened on Sep 15, 2026

[MTP] Use per-module valid-token denominators for AuxLoss

Status: Open.

#4682 In pytorch/torchtitan;

· acisseJZhong opened on Sep 14, 2026

[graph_trainer][pipelining] Carry schedule metadata into reusable stage graphs

Status: Open.

#4655 In pytorch/torchtitan;

· sanketpurandare opened on Sep 14, 2026

[Bug][DistMuon] split fused gated linear weights before NS computation

Status: Open.

#4532 In pytorch/torchtitan;

· WhoisZihan opened on Sep 9, 2026