KD_Lib.Quantization.common

KD_Lib.Quantization.common.base_class module

class KD_Lib.Quantization.common.base_class.Quantizer(model, qconfig, train_loader=None, test_loader=None, optimizer=None, criterion=None, device=device(type='cpu'))

Bases: object

Basic Implementation of Quantization for PyTorch models.

Parameters:

• model (torch.nn.Module) – Model that needs to be pruned
• qconfig (Qconfig) – Configuration used for quantization
• train_loader (torch.utils.data.DataLoader) – DataLoader used for training
• test_loader (torch.utils.data.DataLoader) – DataLoader used for testing
• optimizer (torch.optim.*) – Optimizer for training
• criterion (Loss_fn) – Loss function used for calibration
• device (torch.device) – Device used for training (“cpu” or “cuda”)

get_model_sizes()

Function for printing sizes of the original and quantized model

get_performance_statistics()

Function used for reporting inference performance of original and quantized models Note that performance here referes to the following: 1. Accuracy achieved on the testset 2. Time taken for evaluating on the testset

quantize()

Function used for quantization