Add one pattern to remove the quantize->dequantize pairs for the floating-point constants
The existence of this pattern indicates the user op doesn't have sufficient quantization parameters to be quantized. Then we should keep the floating-point constants as float. PiperOrigin-RevId: 295063207 Change-Id: I759ec06bcda8d15c27834b890dfd41fba3e08d17
Showing
想要评论请 注册 或 登录