Mixed Precision Computing

Post-Training Quantization (PTQ)

Process of converting a pre-trained full-precision model to reduced precision (FP16, INT8, INT4) without retraining, using calibration techniques to determine optimal scale and bias factors.

← Wstecz