mm_audio_tower

third_party/VideoLLaMA2/videollama2/model/__init__.py

@@ -52,9 +52,6 @@ VLLMConfigs = {
 
 
 def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, load_4bit=False, device_map="auto", device="cuda", use_flash_attn=False, **kwargs):
-    if hasattr(model.config, "mm_audio_tower"):
-        nname = model.config.mm_audio_tower.split("/")[-1]
-        model.config.mm_audio_tower = os.path.join(model_path, nname)
     
     if 'token' in kwargs:
         token = kwargs['token']
@@ -209,8 +206,9 @@ def load_pretrained_model(model_path, model_base, model_name, load_8bit=False, l
     else:
         context_len = 2048
 
-    
-
-    
+    if hasattr(model.config, "mm_audio_tower"):
+        nname = model.config.mm_audio_tower.split("/")[-1]
+        model.config.mm_audio_tower = os.path.join(model_path, nname)
+        
 
     return tokenizer, model, processor, context_len