CUDA: use mmvq for mul-mat-id for small batch sizes #18958
+224
−121
Merged
Loading