[Operator Mechanism] Support mm out_dtype for FP16 CUDA (#79549)
* [Operator Mechanism] Support mm out_dtype for FP16 CUDA * [CodeStyle] Refine error messages for FP16 mm out_dtype * [Test] Cover invalid mm input dtype
F
feixi committed
39b7f05d8caa561d647ae00a6e06a23ce7d78ead
Parent: 2a75a2d
Committed by GitHub <noreply@github.com>
on 7/29/2026, 8:52:19 AM