Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

fix incorrect setting for lm-head (#149)

wenhuach219da2beed45dd
diff --git a/examples/language-modeling/main.py b/examples/language-modeling/main.py
index 841c9d2f0..3ef1fb452 100644
--- a/examples/language-modeling/main.py
+++ b/examples/language-modeling/main.py
@@ -290,7 +290,7 @@ if __name__ == '__main__':
f"supported currently")
break
if args.quant_lm_head:
- weight_config[lm_head_layer_name] = {"data_type": "int", "bits": 4, "group_size": 32}
+ weight_config[lm_head_layer_name] = {"data_type": "int"}
transformers_version = [int(item) for item in transformers.__version__.split('.')[:2]]
if transformers_version[0] == 4 and transformers_version[1] < 38:
error_message = "Please upgrade transformers>=4.38.0 to support lm-head quantization."