Added parallelism workaround

app.py

@@ -813,13 +813,16 @@ for param in model.parameters():
 
 eot_token = "<|endoftext|>"
 
-def get_reply(response, history = None, in_stepsize = 2.56, in_horizon_length = 5, in_num_iterations = 10, in_top_k = 2):
+def get_reply(response, username = None, histories = {}, in_stepsize = 2.56, in_horizon_length = 5, in_num_iterations = 10, in_top_k = 2):
+    if username == None or username == "": return "<div class='chatbot'>Enter a username</div>", histories
     stepsize = in_stepsize
     horizon_length = int(in_horizon_length)
     num_iterations = int(in_num_iterations)
     top_k = int(in_top_k)
     if response.endswith(("bye", "Bye", "bye.", "Bye.", "bye!", "Bye!")):
-        return "<div class='chatbot'>Chatbot restarted</div>", None
+        del histories[username]
+        return "<div class='chatbot'>Chatbot restarted</div>", histories
+    history = histories.get(username, None)
     convo_hist = (history if history != None else "How are you?<|endoftext|>") + response + eot_token
     # figure out conditioning text
     tokenized_cond_text = tokenizer.encode(
@@ -874,9 +877,10 @@ def get_reply(response, history = None, in_stepsize = 2.56, in_horizon_length =
             convo_hist = eot_token.join(convo_hist_split)
 
         except:
-            return "<div class='chatbot'>Error occured, chatbot restarted</div>", None
-    
-    return html, convo_hist
+            del histories[username]
+            return "<div class='chatbot'>Error occured, chatbot restarted</div>", histories
+    histories[username] = convo_hist
+    return html, histories
 
 css = """
 .chatbox {display:flex;flex-direction:column}
@@ -889,6 +893,7 @@ css = """
 gr.Interface(fn=get_reply,
              theme="default",
              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)