dialogmodel / README.md
jinymusim's picture
Update README.md
3901994
|
raw
history blame
1.18 kB
metadata
license: mit
datasets:
  - daily_dialog
language:
  - en

Useless ChitChat Language Model

Basic Dialog Model from DialoGPT-small. Finetuned on Daily Dialog dataset.

How to use

Use it as any torch python Language Model

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("gpt")
model = AutoModelForCausalLM.from_pretrained("jinymusim/dialogmodel")

# Take user Input
user_utterance = input('USER> ')
user_utterance = user_utterance.strip()
tokenized_context = tokenizer.encode(user_utterance + tokenizer.eos_token, return_tensors='pt')

# generated a response, limit max_lenght to resonable size 
out_response = model.generate(tokenized_context, 
                                max_length=30,
                                num_beams=2,
                                no_repeat_ngram_size=2,
                                early_stopping=True,
                                pad_token_id=self.tokenizer.eos_token_id)

# Truncate User Input
decoded_response = self.tokenizer.decode(out_response[0], skip_special_tokens=True)[len(user_utterance):]

print(f'SYSTEM> {decoded_response}')