metadata

language: en
tags:
  - en
  - english
  - gpt2
  - gpt3
  - text-generation
  - lm
  - nlp
datasets:
  - cnn_dailymail
widget:
  - text: 'Ever noticed how plane seats appear to be getting smaller and smaller? '
inference:
  parameters:
    max_length: 100
    do_sample: true
    temperature: 0.15

GPT-3 small

Pretrained GPT-3 small, it's architecture intentionally resembles that of GPT-3, model was trained on CNN Daily Mail News dataset for text generation

How to use the model

from transformers import GPT2Tokenizer, GPTNeoForCausalLM

tokenizer = GPT2Tokenizer.from_pretrained('gpt3-small-finetune-cnndaily-news')
model = GPTNeoForCausalLM.from_pretrained('gpt3-small-finetune-cnndaily-news')

text = "Ever noticed how plane seats appear to be getting smaller and smaller? "
input_ids = tokenizer.encode(text, return_tensors='pt')
max_length = 150

sample_outputs = model.generate(input_ids, do_sample=True, max_length=max_length,temperature = 0.15)

for i, sample_output in enumerate(sample_outputs):
    print(">> Generated text {}\n\n{}".format(i+1, tokenizer.decode(sample_output.tolist())))
    print('\n---')

Author

Phan Minh Toan