trunk/b847da23ef2ea009548c60c9d026e6a6ecbe5725: [inductor] Preserve add's alpha when fusing add+mm into addmm (#199811)
- PyTorch: 1956 events in the last 90 days
- PyTorch: 1930th Release in the last 90 days
- Previous: earlier the same day · ciflow/trunk/199871
What happened
Fixes #199698 The description below was generated by an AI assistant (Claude) and has not yet been reviewed by me; this PR is a draft until then. torch.add(inp, a @ b, alpha=0.5) under torch.compile silently computed inp + a @ b . The post-grad patterns that fuse aten.add(inp, aten.mm(a, b)) (both operand orders) into aten.addmm(inp, a, b) don't declare alpha , and _TargetArgsExpr._match drops node kwargs that a pat…
Summary assembled by rule from the sources below