Update app.py

app.py

@@ -138,7 +138,9 @@ def process_image(model, tokenizer, transform, device, check_type, image, text):
         else:
             text_embeds = model.tok_embeddings(input_ids)
 
-        vit_embeds, size1 = model.forward_tokenocr(pixel_values)
+        # vit_embeds, size1 = model.forward_tokenocr(pixel_values)
+        vit_embeds, size1 = model.forward_tokenocr(pixel_values.to(torch.bfloat16))
+
         vit_embeds, size2 = post_process(vit_embeds, target_ratio, check_type)
 
         text_embeds = text_embeds / text_embeds.norm(dim=-1, keepdim=True)