Added parallel workaround

app.py

@@ -8,10 +8,13 @@ tokenizer = BlenderbotTokenizer.from_pretrained(model_name)
 model = BlenderbotForConditionalGeneration.from_pretrained(model_name)
 model.to(device)
 
-def get_reply(response, history = []):
+def get_reply(response, username = None, histories = {}):
+    if username == None or username == "": return "<div class='chatbot'>Enter a username</div>", histories
+    history = histories.get(username, [])
     history.append(response)
     if response.endswith(("bye", "Bye", "bye.", "Bye.")):
-        return "<div class='chatbot'>Chatbot restarted</div>", []
+        history[username] = []
+        return "<div class='chatbot'>Chatbot restarted</div>", histories
     if len(history) > 4: history = history[-4:]
     inputs = tokenizer("    ".join(history), return_tensors="pt")
     inputs.to(device)
@@ -24,8 +27,8 @@ def get_reply(response, history = []):
         cls = "user" if m%2 == 0 else "bot"
         html += "<div class='msg {}'> {}</div>".format(cls, msg)
     html += "</div>"
-    
-    return html, history
+    histories[username] = history
+    return html, histories
 
 css = """
 .chatbox {display:flex;flex-direction:column}
@@ -37,7 +40,8 @@ css = """
 
 gr.Interface(fn=get_reply,
              theme="default",
-             inputs=[gr.inputs.Textbox(placeholder="How are you?"), 
+             inputs=[gr.inputs.Textbox(placeholder="How are you?"),
+                    gr.inputs.Textbox(label="Username"),
                     "state"],
              outputs=["html", "state"],
-             css=css).launch(debug=True, enable_queue=True, share=True)
+             css=css).launch(debug=True, enable_queue=True)