metadata
language:
- en
thumbnail: null
tags:
- conversational
metrics:
- perplexity
DialoGPT model fine-tuned using Amazon's Topical Chat Dataset
This model is fine-tuned from the original DialoGPT-medium.
This model was fine-tuned on a subset of messages from Amazon's Topical Chat dataset (due to processing limitations, I restricted my training data to 50,000 observations.
The dataset spands 8 broad topics and contains conversation partner who do not have defined roles. It was created with the goal of aiding in the effort to build a socialbot that can have deep, engaging open-domain conversations with humans.
The eight broad topics are:
- fashion
- politics
- books
- sports
- general entertainment
- music
- science and technology
- movies
Sample dialogue:
| Role | Response |
|---|---|
| User: | What kind of movies do you like? |
| Bot: | I like marvel movies like iron man |
| User: | Do you like horror films? |
| Bot: | Yea, I like the horror films that have macabre and supernatural themes the most |
| User: | Interesting. |
| Bot: | If you watch a horror movie before you view abstract art, you will enjoy the art more |
| User: | Speaking from experience? |
| Bot: | Yes, I've seen many horror movies where the characters were quite evil and supernatural. |
Using the model
Example code for trying out the model (taken directly from the DialoGPT-medium model card):
from transformers import AutoModelWithLMHead, AutoTokenizer
import torch
tokenizer = AutoTokenizer.from_pretrained("satkinson/DialoGPT-small-marvin")
model = AutoModelWithLMHead.from_pretrained("satkinson/DialoGPT-small-marvin")
# Let's chat for 5 lines
for step in range(5):
# encode the new user input, add the eos_token and return a tensor in Pytorch
new_user_input_ids = tokenizer.encode(input(">> User:") + 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)
# pretty print last ouput tokens from bot
print("DialoGPT: {}".format(tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)))