Issues · pytorch/torchtitan · GitHub

Labels Milestones

DeepEP v2 doesn't work with per-layer compile

#3830

· wwwjn opened

on Jun 30, 2026

RFC CUDAGraphable GraphPP

#3820

· sanketpurandare opened

on Jun 29, 2026

TP + MXFP8 crashing with "mat2 must be col_major, got stride (4096, 1)" error

#3817

· vallet-alexis-col01-sbint opened

on Jun 29, 2026

GraphTrainer DSv3-16B: healthy single-node, MI355 FSDP EP=2 parity gap

module: torch.compile

#3816

· rishisinhanj opened

on Jun 29, 2026

8 GPU Model Tests red on main: gpt-oss PP+FSDP+EP+SACOP MoE _grouped_mm index OOB (CUDA scatter assert == ROCm HSA 0x1016)

#3761

· rishisinhanj opened

on Jun 24, 2026

[RL] Support Qwen3.5 in the RL experiment

#3756

· yichuan-w opened

on Jun 24, 2026

[rl] StickySessionRoutingStrategy can overload a generator for long multi-turn rollouts (no KV-budget awareness / no rebalancing)

#3755

· yichuan-w opened

on Jun 24, 2026

vllm v2 model runner does not work with VLLMGenerator DP routing

#3752

· pzhan9 opened

on Jun 23, 2026

[GraphTrainer] [AutoParallel] Tracker for outstanding issues

#3699

· aditvenk opened

on Jun 17, 2026

[rl] self.gate() aten dispatch to `bmm` instead of `mm`, causing extra batch-invariant `bmm`patching

#3670

· wwwjn opened

on Jun 15, 2026

[rl] vLLM torch.compile (VLLM_COMPILE) on the generator fails: inductor "out variant should have at least one out arg"

#3669

· yichuan-w opened

on Jun 15, 2026

Operator torchtitan.deterministic_topk.default does not have a sharding strategy registered.

#3645

· tianyu-l opened

on Jun 12, 2026