process()

功能说明

进行异常值抑制，无传入值。

调用示例

from modelslim.pytorch.llm_ptq.anti_outlier import AntiOutlier, AntiOutlierConfig
anti_config = AntiOutlierConfig(anti_method="m2")
anti_outlier = AntiOutlier(model, calib_data=dataset_calib, cfg=anti_config, model_type='Llama')
anti_outlier.process() 
calibrator = Calibrator(model, quant_config, calib_data=dataset_calib, disable_level='L0') 
calibrator.run(int_infer=False) 
calibrator.save(qaunt_weight_save_path)