Visible to Intel only — GUID: GUID-624DB501-B618-46E0-A8DE-D1D7FCF01F55
Visible to Intel only — GUID: GUID-624DB501-B618-46E0-A8DE-D1D7FCF01F55
GroupNorm
General
The GroupNorm operation performs the following transformation of the input tensor:
The operation is applied per batch, per group of channels. The gamma and beta coefficients are the optional inputs to the model and need to be specified separately for each channel. The mean and variance are calculated for each group.
Operation attributes
Attribute Name |
Description |
Value Type |
Supported Values |
Required or Optional |
---|---|---|---|---|
Specifies the number of groups G that the channel dimension will be divided into. groups should be divisible by the number of channels |
s64 |
between 1 and the number of channels C in the input tensor |
Required |
|
Indicate whether to output mean and variance which can be later passed to backward op. |
bool |
false , true (default is true) |
Optional |
|
When set to True, this module has inputs gamma and beta |
bool |
false , true (default is true) |
Optional |
|
The constant to improve numerical stability. |
f32 |
Arbitrary positive f32 value, 1e-5 (default) |
Optional |
|
Controls how to interpret the shape of src and dst . |
string |
NCX , NXC (default is NXC ) |
Optional |
Execution arguments
The inputs and outputs must be provided according to below index order when constructing an operation.
Inputs
Index |
Argument Name |
Required or Optional |
---|---|---|
0 |
src |
Required |
1 |
gamma |
Optional |
2 |
beta |
Optional |
Outputs
Index |
Argument Name |
Required or Optional |
---|---|---|
0 |
dst |
Required |
1 |
mean |
Optional |
2 |
variance |
Optional |
Supported data types
GroupNorm operation supports the following data type combinations.
Src / Dst |
Gamma / Beta / Mean / Variance |
---|---|
f32 |
f32 |
bf16 |
f32, bf16 |
f16 |
f32 |