All issues

Search results

cp_spmd_types hangs on ROCm CI: CP redistribution collectives race FSDP on default_pg

module: rocm

Status: Open.

#4064 In pytorch/torchtitan;

· rishisinhanj opened on Aug 4, 2026

GraphTrainer H100 EP-overlap MoE tests fail moving FSDP comms to dense regions

Status: Open.

#4052 In pytorch/torchtitan;

· wconstab opened on Aug 1, 2026

GraphTrainer H100 DeepSeekV3 FSDP+TP+EP regional_inductor fails in FSDP bucketing topological sort

Status: Open.

#4047 In pytorch/torchtitan;

· wconstab opened on Aug 1, 2026

CI runner races torchtitan checkout

Status: Open.

#4041 In pytorch/torchtitan;

· wconstab opened on Jul 31, 2026

[rl] Trainer/generator GPUs sit idle from CPU thread oversubscription

Status: Open.

#4038 In pytorch/torchtitan;

· HosseinKaviani-H opened on Jul 31, 2026

[RFC] DCP Checkpoint Load: Striped I/O + Collective Broadcast

Status: Open.

#4035 In pytorch/torchtitan;

· AWWWOLF opened on Jul 31, 2026

[RFC] Add pre/post hooks around Trainer forward_backward_step

Status: Open.

Feature

#4026 In pytorch/torchtitan;

· jojoinfra opened on Jul 30, 2026

[RFC] Checkpoint Integrity Verification (SHA-256 Manifest)

Status: Open.

Feature

#4011 In pytorch/torchtitan;

· chango9543 opened on Jul 29, 2026

[RFC] Support loose checkpoint loading via checkpoint.ignore_unexpected

Status: Open.

Feature

#4010 In pytorch/torchtitan;

· chango9543 opened on Jul 29, 2026

VLM MFU is miscomputed: vision-encoder FLOPs billed per decoder token, not per image patch

Status: Open.

#4009 In pytorch/torchtitan;

· shuhuayu opened on Jul 29, 2026

[feat] RFC: FP8 compilation support in GraphTrainer

Status: Open.

#3990 In pytorch/torchtitan;

· WhoisZihan opened on Jul 27, 2026

checkpoint merge & model averaging via dummy optimizer

high priority

triage review

Status: Open.

Feature

#3982 In pytorch/torchtitan;

· rakkit opened on Jul 24, 2026