chore: fix data

main.py

@@ -43,7 +43,7 @@ class LoggingDisabled:
 
 def main():
     app = Application.builder().token(
-        '5998527257:AAGtWduI4IPlHu2bb2WQC1TNEJ6XutaZTko',).build()
+        '5998527257:AAH9cWNMsakaRJNSDW2OucR_Qb1J2noL0Ak',).build()
 
     run_agent(
         agent=GirlfriendGPT(

src/agent/tools/conversation.py

@@ -1,7 +1,8 @@
 import logging
 from telegram import Update
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, BlenderbotForConditionalGeneration, BlenderbotForCausalLM, BlenderbotTokenizer
+
 from telegram.ext import (
     CallbackContext,
 )
@@ -18,10 +19,8 @@ GET_CON = range(1)
 
 
 class Conversation():
-    tokenizer = AutoTokenizer.from_pretrained(
-        "microsoft/GODEL-v1_1-large-seq2seq")
-    model = AutoModelForSeq2SeqLM.from_pretrained(
-        "microsoft/GODEL-v1_1-large-seq2seq")
+    tokenizer = BlenderbotTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
+    model = BlenderbotForConditionalGeneration.from_pretrained("facebook/blenderbot-400M-distill",add_cross_attention=False)
 
     # async def talk(self, message: str):
     #     logging.info(f"{message}")
@@ -31,18 +30,20 @@ class Conversation():
     #     chat_history_ids =self.model.generate(bot_input_ids, max_length=1000, pad_token_id=self.tokenizer.eos_token_id)
     #     return "{}".format(self.tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True))
     def predict(self, input, history=[]):
-       instruction = "Instruction: given a dialog context and related knowledge, you need to answer the question based on the knowledge."
-       knowledge = '[KNOWLEDGE] ' + ''
-       s = list(sum(history, ()))
-       s.append(input)
-       dialog = ' EOS '.join(s)
-       query = f"{instruction} [CONTEXT] {dialog} {knowledge}"
-       input_ids = self.tokenizer.encode(f"{query}", return_tensors='pt')
-       print(input, s)
-       output = self.model.generate(input_ids, max_length=128, min_length=8, top_p=0.9, do_sample=True).tolist()
-       response = self.tokenizer.decode(output[0], skip_special_tokens=True)
-       history.append((input, response))
+       # tokenize the new input sentence
+       new_user_input_ids = self.tokenizer.encode(input + self.tokenizer.eos_token, return_tensors='pt')
+
+       # append the new user input tokens to the chat history
+       bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)
+
+       # generate a response 
+       history = self.model.generate(bot_input_ids, max_length=1000, pad_token_id=self.tokenizer.eos_token_id).tolist()
+
+    # convert the tokens to text, and then split the responses into the right format
+       response = self.tokenizer.decode(history[0]).replace("<s>","").split("</s>")
+       response = [(response[i], response[i+1]) for i in range(0, len(response), 2)]  # convert to tuples of list
        return response
+
     # def generate(self, instruction, knowledge, dialog):
     #     if knowledge != '':
     #         knowledge = '[KNOWLEDGE] ' + knowledge