OpenGVLab
/

VideoChat-Flash-Qwen2-7B_res224

Video-Text-to-Text

videochat_flash_qwen

feature-extraction

Eval Results (legacy)

Model card Files Files and versions

lixinhao commited on Jan 13, 2025

Commit

156a50d

·

verified ·

1 Parent(s): a6cbec7

Update modeling_videochat_flash.py

Files changed (1) hide show

modeling_videochat_flash.py +1 -1

modeling_videochat_flash.py CHANGED Viewed

@@ -641,7 +641,7 @@ class VideoChatFlashQwenForCausalLM(LlavaMetaForCausalLM, Qwen2ForCausalLM_Flash
         conv = conv_templates["qwen_2"].copy()
         if chat_history is None or len(chat_history) == 0:
-            user_prompt = f'{DEFAULT_IMAGE_TOKEN}\n{time_msg.rstrip()} {user_prompt}'
         else:
             assert DEFAULT_IMAGE_TOKEN in chat_history[0]['content'], chat_history
             for msg in chat_history:

         conv = conv_templates["qwen_2"].copy()
         if chat_history is None or len(chat_history) == 0:
+            user_prompt = f'{DEFAULT_IMAGE_TOKEN}\n{time_msg.strip()} {user_prompt}'
         else:
             assert DEFAULT_IMAGE_TOKEN in chat_history[0]['content'], chat_history
             for msg in chat_history: