Search results

After applying commit 6a3a9da, the training job runs out of GPU memory (OOM).

Status: Open.

Bug

#1855 In pytorch/torchtitan;

· limou102 opened on Oct 11, 2025

Pipeline Parallelism Across Nodes Fails with EOFError

Status: Open.

Bug

#1852 In pytorch/torchtitan;

· rrutmann opened on Oct 10, 2025

More robust distributed loss

high priority

triage review

Status: Open.

Feature

#1842 In pytorch/torchtitan;

· lkhphuc opened on Oct 9, 2025

TP gradient update is wrong during MoE backward

Status: Open.

#1805 In pytorch/torchtitan;

· wwwjn opened on Oct 7, 2025

Deepseekv3 CUDA illegal memory access on B200 from increasing seq_len to 8192

bugSomething isn't workingSomething isn't working

Status: Open.

Bug

#1791 In pytorch/torchtitan;

· danielvegamyhre opened on Oct 3, 2025

How to add supervised finetuning mask in torchtitan?

post training

Status: Open.

Feature

#1781 In pytorch/torchtitan;

· kailashg26 opened on Oct 1, 2025

Posttraining Library

Status: Open.

#1771 In pytorch/torchtitan;

· MarkLiLabs opened on Sep 30, 2025

unwaited collective in DSV3+FSDP2+TP+compile

module: torch.compile

Status: Open.

Bug

#1769 In pytorch/torchtitan;

· ruisizhang123 opened on Sep 29, 2025

Async TP CI failing

bugSomething isn't workingSomething isn't working

ci test failure

Status: Open.

Bug

#1757 In pytorch/torchtitan;

· fegin opened on Sep 25, 2025

Inconsistent loss between different TP

questionFurther information is requestedFurther information is requested

Status: Open.

#1750 In pytorch/torchtitan;

· weixuansun opened on Sep 24, 2025

What is the benefit of using torchrun instead of python directly with slurm and other launchers ?

Status: Open.

#1749 In pytorch/torchtitan;

· githubsgi opened on Sep 23, 2025

Adding options to enable some determinism related configs

Status: Open.

#1736 In pytorch/torchtitan;

· githubsgi opened on Sep 22, 2025