Chris Staff asked 1 month ago I am implementing DialoGPT with HuggingFace Transformers, like this:
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
def load_tokenizer_and_model(model="microsoft/DialoGPT-small"):
"""
Load tokenizer and model instance for some specific DialoGPT model.
"""
# Initialize tokenizer and model
print("Loading model...")
tokenizer = AutoTokenizer.from_pretrained(model)
model = AutoModelForCausalLM.from_pretrained(model)
# Return tokenizer and model
return tokenizer, model
def generate_response(tokenizer, model, step):
"""
Generate a response to some user input.
"""
# encode the new user input, add the eos_token and return a tensor in Pytorch
new_user_input_ids = tokenizer.encode(input(">> You:") + tokenizer.eos_token, return_tensors='pt')
# append the new user input tokens to the chat history
bot_input_ids = torch.cat([chat_history_ids, new_user_input_ids], dim=-1) if step > 0 else new_user_input_ids
# generated a response while limiting the total chat history to 1000 tokens,
chat_history_ids = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id)
# Print response
print("DialoGPT: {}".format(tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)))
def chat_for_n_rounds(n=5):
"""
Chat with chatbot for n rounds (n = 5 by default)
"""
# Initialize tokenizer and model
tokenizer, model = load_tokenizer_and_model()
# Chat for n rounds
for step in range(n):
generate_response(tokenizer, model, step)
if __name__ == '__main__':
chat_for_n_rounds(5)
However, I am getting the following error:
Traceback (most recent call last):
File "dialogpt.py", line 48, in
if __name__ == '__main__':
File "dialogpt.py", line 45, in chat_for_n_rounds
for step in range(n):
File "dialogpt.py", line 26, in generate_response
# append the new user input tokens to the chat history
UnboundLocalError: local variable 'chat_history_ids' referenced before assignment
Let's pause for a second! 👩💻
We send emails at least every Friday. Welcome!
What can be wrong here?
Chris Staff answered 1 month ago The issue lies within generate_response, in the sense that you are using chat_history_ids before it is assigned. This always occurs during the first run, because a chat history is not available, but you still need it for concatenation.
To solve this, you must make available the chat_history_ids variable to the outer loop – pass it to generate_response and also return it – like this:
def generate_response(tokenizer, model, chat_round, chat_history_ids):
"""
Generate a response to some user input.
"""
# Encode user input and End-of-String (EOS) token
new_input_ids = tokenizer.encode(input(">> You:") + tokenizer.eos_token, return_tensors='pt')
# Append tokens to chat history
bot_input_ids = torch.cat([chat_history_ids, new_input_ids], dim=-1) if chat_round > 0 else new_input_ids
# Generate response given maximum chat length history of 1250 tokens
chat_history_ids = model.generate(bot_input_ids, max_length=1250, pad_token_id=tokenizer.eos_token_id)
# Print response
print("DialoGPT: {}".format(tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)))
# Return the chat history ids
return chat_history_ids
This allows you to re-use the chat history ids in chat_for_n_rounds and assign None to it during the first round, resolving the error – like this:
def chat_for_n_rounds(n=5):
"""
Chat with chatbot for n rounds (n = 5 by default)
"""
# Initialize tokenizer and model
tokenizer, model = load_tokenizer_and_model()
# Initialize history variable
chat_history_ids = None
# Chat for n rounds
for chat_round in range(n):
chat_history_ids = generate_response(tokenizer, model, chat_round, chat_history_ids)
