Visible to Intel only — GUID: GUID-E9E5EC3E-14F7-4715-A087-D3EE12C65653
Visible to Intel only — GUID: GUID-E9E5EC3E-14F7-4715-A087-D3EE12C65653
Quantize
General
Quantize operation converts an f32 tensor to a quantized (u8/s8/f8_e4m3/f8_e5m2) tensor. It supports both per-tensor and per-channel asymmetric linear quantization. Output data type is specified in the output tensor data type. Rounding mode is library-implementation defined. Zero points (zps in the attribute table) are not supported for f8_e4m3 and f8_e5m2 quantization.
For per-tensor quantization:
For per-channel quantization, taking channel axis = 1 as an example:
where is the number of channels.
Operation attributes
Attribute Name |
Description |
Value Type |
Supported Values |
Required or Optional |
---|---|---|---|---|
Specifies which quantization type is used. |
string |
per_tensor (default), per_channel |
Optional |
|
Specifies dimension on which per-channel quantization is applied. |
s64 |
A s64 value in the range of [-r, r-1] where r = rank(src), 1 by default |
Optional |
|
Scalings applied on the src data. |
f32 |
A f32 list (only contain one element if qtype is per_tensor ) |
Required |
|
Offset values that maps to float zero. |
s64 |
A s64 list (only contain one element if qtype is per_tensor ). If omitted, zps values are assumed to be zero. |
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 |
Outputs
Index |
Argument Name |
Required or Optional |
---|---|---|
0 |
dst |
Required |
Supported data types
Quantize operation supports the following data type combinations.
Src |
Dst |
---|---|
f32 |
s8, u8, f8_e4m3, f8_e5m2 |