disable load_in_4bit=True

configuration_InternLM_XComposer.py

@@ -35,7 +35,7 @@ class InternLMXComposerConfig(PretrainedConfig):
         kqvo_bias=True,
         device='cuda',
         internlm_lora=None,
-        load_in_4bit=True,
+        # load_in_4bit=True,
         **kwargs,
     ):
         self.vocab_size = vocab_size