PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
refine example (#224)
WeiweiZhang1ac77f1e5116d
diff --git a/examples/language-modeling/main.py b/examples/language-modeling/main.pyindex d476fd948..7162b1a91 100644--- a/examples/language-modeling/main.py+++ b/examples/language-modeling/main.py@@ -135,7 +135,7 @@ if __name__ == '__main__':parser.add_argument("--act_bits", default=32, type=int,help="activation bits")- parser.add_argument("--fp_layers_list", default="", type=str,+ parser.add_argument("--fp_layers", default="", type=str,help="List of Layers to maintain original data type")args = parser.parse_args()@@ -275,12 +275,12 @@ if __name__ == '__main__':layer_config[n] = {"bits": 32}print(f"{n} will not be quantized due to its shape not being divisible by 32, resulting in an exporting issue to autogptq")- fp_layers_list = args.fp_layers_list.split(",")- if bool(fp_layers_list):+ fp_layers = args.fp_layers.split(",")+ if bool(fp_layers):for n, m in model.named_modules():if isinstance(m, torch.nn.Linear) or isinstance(m, transformers.modeling_utils.Conv1D):name = n.split('.')[-1]- if n in fp_layers_list or name in fp_layers_list:+ if n in fp_layers or name in fp_layers:layer_config[n] = {"bits": 32}print(f"{n} will not be quantized.")