Interface Description

Table 1 lists the adapted APIs supported by KuDNN.

**Table 1** Interfaces
Operator	Function	Data Type Supported by KuDNN
Linear	Linear transformation	input weight bias output (fp16 fp16 fp16 fp16) input weight bias output (fp16 fp16 fp32 fp32) input weight bias output (int8 int8 int32 int32) input weight bias output (fp16 int8 int32 fp32) input weight bias output (fp32 fp32 fp32 fp32)
SoftMax	Probability distribution normalization	fp32
Conv	Local feature extraction	fp16, fp32
GroupNorm	Channel group normalization	fp16, fp32
LayerNorm	Layer group normalization	fp16, fp32
RMSNorm	Root mean square normalization	fp16
Embedding	Maps integer indices to fixed vectors	fp16

Parent topic: Operator Interfaces