Update app.py

app.py

@@ -55,7 +55,9 @@ tokenizer, model, image_processor, context_len = load_pretrained_model(
 model.config.training = False
 
 # FIXED PROMPT
-FIXED_PROMPT = "<image>\nWhat material is this item and how to dispose of it?"
+FIXED_PROMPT = "What type of waste is this item and how to dispose of it?"
+INTERNAL_PROMPT = "<image>\nWhat type of waste is this item and how to dispose of it?"
+
 
 def clear_history():
     state = default_conversation.copy()
@@ -66,13 +68,15 @@ def add_text(state, imagebox, textbox, image_process_mode):
         state = conv_templates[conv_mode].copy()
 
     if imagebox is not None:
-        textbox = FIXED_PROMPT
         image = Image.open(imagebox).convert('RGB')
-        textbox = (textbox, image, image_process_mode)
-    state.append_message(state.roles[0], textbox)
-    state.append_message(state.roles[1], None)
+        visible_text = FIXED_PROMPT
+        model_prompt = (INTERNAL_PROMPT, image, image_process_mode)
+        state.append_message(state.roles[0], visible_text)
+        state.append_message(state.roles[1], None)
+        state.messages[-2].append(model_prompt)  
     yield (state, state.to_gradio_chatbot(), "", None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
 
+
 def delete_text(state, image_process_mode):
     state.messages[-1][-1] = None
     prev_human_msg = state.messages[-2]
@@ -90,7 +94,11 @@ def regenerate(state, image_process_mode):
 
 @spaces.GPU
 def generate(state, imagebox, textbox, image_process_mode, temperature, top_p, max_output_tokens):
-    prompt = FIXED_PROMPT
+    hidden_data = state.messages[-2]
+    if len(hidden_data) == 3 and isinstance(hidden_data[2], tuple):
+        prompt, image, image_process_mode = hidden_data[2]
+    else:
+        prompt = FIXED_PROMPTT
     images = state.get_images(return_pil=True)
 
     ori_prompt = prompt