Choose how the model outputs prediction results: quantized (integer) or floating point. The following options are available:
- Quantized 8-bit — Probabilities from 0 to 255.
- Quantized 16-bit — Probabilities from 0 to 65,535.
- Floating-point 32-bit — Probabilities from 0 to 1.
Quantized format is only available when quantized weights and coefficients are selected. For 32-bit (floating-point) weights, only floating-point output is available and cannot be changed.
Note
When the LiteRT framework is used, the value is predefined to 32-bit probabilities and cannot be changed.