- trunk/f3d603baf74daa6ac8fbf15feb37f0edb6ac9ad3: Avoid re-inferring DeepSeek recipes in the TMA heuristic (#197458)
PyTorch Releases
- trunk/599f186d3ff3a0e124544ecd62ca63c0a285c32d: [GPU][AOTI] Support host TMA in C++ wrappers (#193608) (#193608)
PyTorch Releases
- viable/strict/1788418752: [GPU][AOTI] Support host TMA in C++ wrappers (#193608) (#193608)
PyTorch Releases