onnxruntime
QAttention calls into MatMulIntToFloat instead of Dequantize+GEMM
#16851
Merged

Loading