[Torchtitan][Pytorch][ROCm 7.14] Qwen3 MoE TP+EP training produces extremely large(nan) gradient norms

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

FlexAttention `BlockMask` breaks Pipeline Parallelism with split-backward (zero-bubble / multi) schedules

inexistent etp flag in the experiments/ folder

CI: 8 GPU Integration Test on H100 ROCm leg red on main (blocked on pytorch/test-infra#8152)

[RL][Flex] batch invariant Flex Attention doesn't work with general compile

HybridEP DispatchHandle not working with torch.compile

[gpt_oss] Avoid materialising `(x_linear + 1)` intermediate in `swiglu` to reduce activation memory

DeepEP: CUDA stream race between masked_fill and get_dispatch_layout causes grouped_mm assertion failure

torch.compile fails with MoE + EP: PendingUnbackedSymbolNotFound on DTensor .view(-1, dim)