NVFP4 kernel pipeline gets PDL, fused scaling, and tuned autotuning; torchtitan drops stale train command
- [inductor][NVGEMM] Chain dependent NVFP4 launches into Triton consumers (#197485) pytorch/pytorch
- [inductor][NVGEMM] Fuse NVFP4 output scaling (#194651) pytorch/pytorch
- [inductor][NVGEMM] Improve NVFP4 autotuning (#194652) pytorch/pytorch
- [inductor] Make CUDA graph autotuning measurements consistent (#194649) pytorch/pytorch