Versioned name: DynamicDequantize-1
Short description: DynamicDequantize converts a quantized (s8 or u8) tensor to a f32 tensor. It supports both per-tensor and per-channel asymmetric linear de-quantization. Rounding mode is library-implementation defined. Unlike the static version of Dequantize, DynamicDequantize takes scales and zero-points as operator input tensors.
Description: specifies which de-quantization type is used.
Range of values: “per_tensor” or “per_channel”
Default value: “per_tensor”
Description: specifies the dimension on which “per-channel” de-quantization is applied. The attributes is valid only when qtype is “per_channel”.
Range of values: integers in [-r, r-1] where r = rank(input). Negative value means counting the dimension backwards from the end.
Default value: 1.
input- s8/u8 tensor to be de-quantized. Required.
scales- f32 1D tensor to be applied to the quantization formula. For qtype = per-tensor, there should be only one element in the scales tensor. For qtype = per-channel, the element number should be equal to the element number of input tensor along the dimension axis. Required.
zps- u8/s8/s32 1D tensor with offset values that map to zero. For qtype = per-tensor, there should be only one element in the zps tensor. For qtype = per-channel, the element number should be equal to the element number of input tensor along the dimension axis. If not specified, the library can assume the operator is symmetric de-quantization and perform kernel optimization accordingly. Optional.
output- f32 de-quantized tensor.
T1: s8, u8.
T3: s8, u8, s32.