lixinhao commited on
Commit
10a3b5a
·
verified ·
1 Parent(s): c7a5c8f

Update modeling_videochat_flash.py

Browse files
Files changed (1) hide show
  1. modeling_videochat_flash.py +2 -2
modeling_videochat_flash.py CHANGED
@@ -636,7 +636,7 @@ class VideoChatFlashQwenForCausalLM(LlavaMetaForCausalLM, Qwen2ForCausalLM_Flash
636
 
637
  image_sizes = [frames[0].shape[:2]]
638
 
639
- frames = [self.get_vision_tower().image_processor.preprocess(frames, return_tensors="pt")["pixel_values"].half().cuda()]
640
 
641
  conv = conv_templates["qwen_2"].copy()
642
 
@@ -679,7 +679,7 @@ class VideoChatFlashQwenForCausalLM(LlavaMetaForCausalLM, Qwen2ForCausalLM_Flash
679
 
680
  outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
681
  if outputs.endswith(stop_str):
682
- outputs = outputs[: -len(stop_str)]
683
 
684
  outputs = outputs.strip()
685
 
 
636
 
637
  image_sizes = [frames[0].shape[:2]]
638
 
639
+ frames = [self.get_vision_tower().image_processor.preprocess(frames, return_tensors="pt")["pixel_values"].to(self.model.dtype).cuda()]
640
 
641
  conv = conv_templates["qwen_2"].copy()
642
 
 
679
 
680
  outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
681
  if outputs.endswith(stop_str):
682
+ outputs = outputs[: -len(stop_str)]
683
 
684
  outputs = outputs.strip()
685