onnxruntime
Quantization tool: support float 8 with MatMul, support float 16 weights
#18043
Merged
Merged