Stateful model -> memory

app.py

@@ -7,18 +7,24 @@ from transformers import GPT2LMHeadModel, GPT2Tokenizer
 tokenizer = GPT2Tokenizer.from_pretrained('microsoft/DialoGPT-medium')
 model = GPT2LMHeadModel.from_pretrained('microsoft/DialoGPT-medium')
 
-def chat(message, token_response):
-    token_message = tokenizer.encode(message + tokenizer.eos_token, return_tensors='pt')
-    token_response = model.generate(token_message, max_length=1000, pad_token_id=tokenizer.eos_token_id)
-    response = tokenizer.decode(token_response[:, token_message.shape[-1]:][0], skip_special_tokens=True)
-    return response, token_response
+def predict(input, history=[]):
+    # tokenize the new input sentence
+    new_user_input_ids = tokenizer.encode(input + tokenizer.eos_token, return_tensors='pt')
 
-input = gr.inputs.Textbox(lines=2, label='User:')
-output = gr.outputs.Textbox(label='Bot:')
+    # append the new user input tokens to the chat history
+    bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)
 
-gr.Interface(fn=chat,
+    # generate a response 
+    history = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id).tolist()
+
+    # convert the tokens to text, and then split the responses into lines
+    response = tokenizer.decode(history[0]).split("<|endoftext|>")
+    response = [(response[i], response[i+1]) for i in range(0, len(response)-1, 2)]  # convert to tuples of list
+    return response, history
+
+gr.Interface(fn=predict,
     title="DialoGPT-medium",
-    inputs=[input, "state"],
-    outputs=[output, "state"],
+    inputs=["text", "state"],
+    outputs=["text", "state"],
     allow_screenshot=False,
     allow_flagging='never').launch()