adapt dimension of blip inputs

app.py

@@ -45,12 +45,11 @@ if uploaded_file is not None:
     pixel_values = blip_processor(images=img, return_tensors="pt").pixel_values
     blip_ques = blip_processor.tokenizer.cls_token + question
     batch_input_ids = blip_processor(text=blip_ques, add_special_tokens=False).input_ids
-    batch_input_ids = torch.tensor(batch_input_ids)
-    st.text(f"input dimension: {batch_input_ids.shape}")
-    st.text(f"pixel dimension: {pixel_values.shape}")
-    # generate_ids = blip_model.generate(pixel_values=pixel_values, input_ids=batch_input_ids, max_length=50)
-    # blip_output = blip_processor.batch_decode(generate_ids, skip_special_tokens=True)
+    batch_input_ids = torch.tensor(batch_input_ids).unsqueeze(0)
+
+    generate_ids = blip_model.generate(pixel_values=pixel_values, input_ids=batch_input_ids, max_length=50)
+    blip_output = blip_processor.batch_decode(generate_ids, skip_special_tokens=True)
 
     st.text(f"Answer of ViLT: {orig_pred}")
     st.text(f"Answer after fine-tuning: {pred}")
-    # st.text(f"Answer of BLIP: {blip_output}")
+    st.text(f"Answer of BLIP: {blip_output[0]}")