Fix torch awq (#4503)

* fix glm4.7-flash

* fix awq

Author: grimoire

lmdeploy/pytorch/config.py

@@ -631,6 +631,9 @@ def from_config(cls, hf_config: Any):
         if quant_method == 'awq':
             bits = quant_config.get('bits', 4)
             group_size = quant_config.get('group_size', 128)
+            if quant_dtype is None:
+                # awq does not need a quant dtype, this is just a placeholder
+                quant_dtype = 'bfloat16'
         elif quant_method == 'smooth_quant':
             if quant_dtype is None:
                 quant_dtype = 'int8'

lmdeploy/pytorch/models/utils/model.py

@@ -80,6 +80,8 @@ def update_quant_config(cls, quant_config: QuantizationConfig):
         """Update quant config."""
         if quant_config is None:
             return
+        if getattr(quant_config, 'ignored_layers', None) is None:
+            return quant_config
         ignored_layers = [cls.rename_weight(name) for name in quant_config.ignored_layers]
 
         added_ignore_layers = set()