[fix] fix image ilen

modeling_videochat2.py

@@ -257,7 +257,7 @@ class InternVideo2_VideoChat2(BaseMLLM):
       return_history =False,
       generation_config={}
     ):
-        ilen = media_tensor.shape[1]
+        
 
         conversation = ""
         if instruction:
@@ -268,8 +268,10 @@ class InternVideo2_VideoChat2(BaseMLLM):
                 )
 
         if media_type == 'image':
+            ilen = media_tensor.shape[0]
             conversation +=( "<img>" + IMG_TOKEN + "</img>")*ilen
         else:
+            ilen = media_tensor.shape[1]
             conversation += ("<vid>" + VID_TOKEN + "</vid>")*ilen