Search results

Triton Cache Error for MI250

Status: Open.

Bug

#2146 In pytorch/torchtitan;

· zixianwang2022 opened on Dec 11, 2025

Validation breaking with FlashAttention

Status: Open.

#2140 In pytorch/torchtitan;

· francesco-bertolotti opened on Dec 11, 2025

TorchComms backend (NCCLX and TorchComms NCCL) initializes but is ignored / not bound to PG in Torchtitan FSDP + DeviceMesh pipeline → large model hangs

Status: Open.

Bug

#2139 In pytorch/torchtitan;

· sk2693 opened on Dec 10, 2025

Missing model.safetensors.index.json in final hf checkpoint

Status: Open.

#2137 In pytorch/torchtitan;

· legarwin opened on Dec 10, 2025

Duplicated AG/RS issued from Simple FSDP if there are multiple reference to the same parameters

module: torch.compile

Status: Open.

Bug

#2133 In pytorch/torchtitan;

· yanboliang opened on Dec 10, 2025

precommit run fails to install all dependencies on 3.14

Status: Open.

Bug

#2126 In pytorch/torchtitan;

· divyanshk opened on Dec 9, 2025

Knowledge Distillation template

questionFurther information is requestedFurther information is requested

Status: Open.

#2109 In pytorch/torchtitan;

· Separius opened on Dec 4, 2025

EP in latest main is slow

Status: Open.

#2101 In pytorch/torchtitan;

· goldhuang opened on Dec 3, 2025

symmetric memory all2all integration for EP

Status: Open.

#2100 In pytorch/torchtitan;

· goldhuang opened on Dec 3, 2025

Torchtitan CI gap Between ROCm & CUDA

module: rocm

Status: Open.

Task

#2098 In pytorch/torchtitan;

· akashveramd opened on Dec 2, 2025

[Possible PR discuss] Automatic Pipeline Parallel Partitioning Considering Embedding Layers

Status: Open.

Feature

#2093 In pytorch/torchtitan;

· McmillanTAC opened on Dec 1, 2025

save optimizer state with HF

Status: Open.

#2090 In pytorch/torchtitan;

· silingtong123 opened on Nov 28, 2025