Search results

Check tensor parallelism degree in parallelize_fn

Status: Open.

Task

#2574 In pytorch/torchtitan;

· gali-leilei opened on Mar 14, 2026

RoPE positions are never set

Status: Open.

Bug

#2559 In pytorch/torchtitan;

· francesco-bertolotti opened on Mar 12, 2026

[RFC] Support new models Qwen3.5-35B-A3B (text input only)

Status: Open.

Feature

#2544 In pytorch/torchtitan;

· gali-leilei opened on Mar 11, 2026

[RFC] Magi Attention: Varlen + Context Parallelism with Document-Aware Load Balancing

Status: Open.

Feature

#2536 In pytorch/torchtitan;

· wangbinluo opened on Mar 10, 2026

Adding Z-Loss

Status: Open.

Feature

#2523 In pytorch/torchtitan;

· francesco-bertolotti opened on Mar 7, 2026

[Regression tracking] CPU offload hangs forever in torchtitan CI

Status: Open.

Bug

#2518 In pytorch/torchtitan;

· wwwjn opened on Mar 6, 2026

Support LoRA at TorchTitan

Status: Open.

#2515 In pytorch/torchtitan;

· mori360 opened on Mar 6, 2026

·  

New Feature, Model, Misc

[Compiler Toolkit] Using copy-engine collectives for prefetching

module: torch.compile

Status: Open.

#2514 In pytorch/torchtitan;

· xmfan opened on Mar 6, 2026

·  

New Feature, Model, Misc

[RL] Improve vLLM/Generator startup time with cudagraphs and support_torch_compile

Status: Open.

Task

#2509 In pytorch/torchtitan;

· Lucaskabela opened on Mar 6, 2026

[RL][Feature Request] Tun on torch.compile + cudagraphs for trainer definition

Status: Open.

Feature

#2508 In pytorch/torchtitan;

· Lucaskabela opened on Mar 6, 2026

Llama4 debug_model with DeepEP - ValueError: not enough values to unpack (expected 5, got 2)

Status: Open.

Bug

#2505 In pytorch/torchtitan;

· githubsgi opened on Mar 5, 2026

[PyTorch Regression] FlexAttention with wrap_inductor_compiled_regions doesn't work

bugSomething isn't workingSomething isn't working

high priority

triage review

Status: Open.

#2463 In pytorch/torchtitan;

· fegin opened on Feb 28, 2026

·  

Bugs, Composability