`neural_compressor.adaptor.torch_utils.awq`

Module Contents

Implementation of Activation-aware Weight quantization (AWQ) algo.

class neural_compressor.adaptor.torch_utils.awq.ActAwareWeightQuant(model, example_inputs=None, calib_func=None, dataloader=None, n_samples=128, data_type='int', bits=4, group_size=32, scheme='asym', enable_full_range=False, weight_config={})[source]: Implementation of Activation-aware Weight quantization (AWQ) algo.