Update app.py

app.py

@@ -88,14 +88,21 @@ def bot_streaming(message, history):
     image = Image.open(image).convert('RGB')
     image_tensor = process_images([image], image_processor, model.config)[0]
 
+    
+
     inputs = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0)
 
-    print('inputs', inputs)
+    
+    image_tensor = image_tensor.to(dtype=torch.float16, device='cuda', non_blocking=True)
+    input_ids.to(device='cuda', non_blocking=True)
+
+    print('image', image_tensor.shape)
+    print('inputs', inputs.shape)
 
     
     streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
     
-    generation_kwargs = dict(input_ids=inputs, streamer=streamer, max_new_tokens=1024, do_sample=False, eos_token_id = terminators)
+    generation_kwargs = dict(input_ids=inputs, images=image_tensor, streamer=streamer, max_new_tokens=1024, do_sample=False, eos_token_id = terminators)
 
     print('here?')