update

app.py

@@ -49,14 +49,16 @@ def main(args):
     
     title = """<h1 align="center">Demo of Tiger</h1>"""
     description1 = """<h2>This is the demo of Tiger (Generative Multimodal Dialogue Model).</h2>"""
-    description2 = """<h3>Input text start chatting!</h3>"""
-    description_input = """<h3>Input:  text</h3>"""
-    description_output = """<h3>Output: text / image</h3>"""
+    description2 = """<h2>Input text start chatting!</h2>"""
+    hr = """<hr>"""
+    description_input = """<h3>Input:  text (English)</h3>"""
+    description_output = """<h3>Output:  text / image</h3>"""
 
     with gr.Blocks() as demo:
         gr.Markdown(title)
         gr.Markdown(description1)
         gr.Markdown(description2)
+        gr.Markdown(hr)
         gr.Markdown(description_input)
         gr.Markdown(description_output)
 
@@ -78,15 +80,17 @@ def main(args):
                     interactive=True,
                     label="seed for text-to-image",
                 )
-                clear = gr.Button("Restart (Clear dialogue history)")
+                start = gr.Button("Start Chat", variant="primary")
+                clear = gr.Button("Restart Chat (Clear dialogue history)", interactive=False)
 
             with gr.Column():
                 chat_state = gr.State()
                 chatbot = gr.Chatbot(label='Tiger')
-                text_input = gr.Textbox(label='User', placeholder='Please input the text.')
-
+                text_input = gr.Textbox(label='User', placeholder="Please click the <Start Chat> button to start chat!", interactive=False)
+        
+        start.click(chat.start_chat, [chat_state], [text_input, start, clear, chat_state])
         text_input.submit(chat.respond, [text_input, num_beams, text2image_seed, chatbot, chat_state], [text_input, chatbot, chat_state])
-        clear.click(lambda: None, None, chatbot, queue=False)
+        clear.click(chat.restart_chat, [chat_state], [chatbot, text_input, start, clear, chat_state], queue=False)
 
     demo.launch(share=False, enable_queue=False)
 

chatbot.py

@@ -1,5 +1,7 @@
 import os
 import sys
+import gradio as gr
+from datetime import datetime
 
 import torch
 from model import IntentPredictModel
@@ -34,8 +36,21 @@ class Chat:
         
         self.context_for_intent = ""
         self.context_for_text_dialog = ""
+        
+    def start_chat(self, chat_state):
+        self.context_for_intent = ""
+        self.context_for_text_dialog = ""
+        
+        return gr.update(interactive=True, placeholder='input the text (English).'), gr.update(value="Start Chat", interactive=False), gr.update(value="Restart Chat (Clear dialogue history)", interactive=True), chat_state
+    
+    def restart_chat(self, chat_state):
+        self.context_for_intent = ""
+        self.context_for_text_dialog = ""
+        
+        return None, gr.update(interactive=False, placeholder='Please click the "Start Chat" button.'), gr.update(value="Start Chat", interactive=True), gr.update(value="Restart Chat (Clear dialogue history)", interactive=False), chat_state
     
     def intent_predict(self, context: str):
+        print(f"context = {context}")
         context_encoded = self.intent_predict_tokenizer.encode_plus(
             text=context,
             add_special_tokens=True,
@@ -67,7 +82,7 @@ class Chat:
         
         generated_ids = self.text_dialog_model.generate(input_ids.to(self.device),
                                                         max_new_tokens=64, min_new_tokens=3,
-                                                        do_sample=False, num_beams=num_beams, length_penalty=0.7, num_beam_groups=5, 
+                                                        do_sample=False, num_beams=num_beams, length_penalty=0.7, num_beam_groups=num_beams, 
                                                         no_repeat_ngram_size=3,
                                                         bad_words_ids=tokenizer(bad_words, add_prefix_space=True, add_special_tokens=False).input_ids,
                                                         forced_decoder_ids=[[input_ids.shape[-1], tag_id_dic[tag]]],  # 指定生成的回复中第一个token始终是tag(因为generated_ids中包括input_ids, 所以是第input_ids.shape[-1]位)  
@@ -90,6 +105,9 @@ class Chat:
         return response_str
     
     def respond(self, message, num_beams, text2image_seed, chat_history, chat_state):
+        current_time = datetime.now().strftime("%b%d_%H-%M-%S")
+        print("=" * 50)
+        print(f"Time: {current_time}")
         print(f"User: {message}")
         # process context
         if self.context_for_intent == "":
@@ -124,7 +142,7 @@ class Chat:
             self.context_for_intent += " [SEP] " + response
             self.context_for_text_dialog += "[DST] " + response
             
-            chat_history.append((message, (save_image_path, None)))
+            chat_history.append((message, (save_image_path, f"Generated Image Caption: {response}")))
 
         else:
             print(f"Bot: {response}")