OpenGVLab
/

VideoChat-Flash-Qwen2_5-2B_res448

Video-Text-to-Text

videochat_flash_qwen

feature-extraction

Model card Files Files and versions Community

lixinhao commited on Mar 4

Commit

6e39ace

·

verified ·

1 Parent(s): aac07d6

Update modeling_videochat_flash.py

Files changed (1) hide show

modeling_videochat_flash.py +1 -1

modeling_videochat_flash.py CHANGED Viewed

@@ -636,7 +636,7 @@ class VideoChatFlashQwenForCausalLM(LlavaMetaForCausalLM, Qwen2ForCausalLM_Flash
         image_sizes = [frames[0].shape[:2]]
-        frames = [self.get_vision_tower().image_processor.preprocess(frames, return_tensors="pt")["pixel_values"].half().cuda()]
         conv = conv_templates["qwen_2"].copy()

         image_sizes = [frames[0].shape[:2]]
+        frames = [self.get_vision_tower().image_processor.preprocess(frames, return_tensors="pt")["pixel_values"].to(self.model.dtype).cuda()]
         conv = conv_templates["qwen_2"].copy()