gpt

[ghost]

Thank you

[enochlev]

Heres a start

!pip install transformers
import transformers
config = transformers.GPT2Config.from_pretrained('/content/checkpoint/run3/hparams.json')
tokenizer = transformers.GPT2Tokenizer("checkpoint/run3/encoder.json", "checkpoint/run3/vocab.bpe")
model = transformers.GPT2Model.from_pretrained('/content/checkpoint/run3/model-1500.index',from_tf=True,config=config)

model.save_pretrained('gpt2_')
tokenizer.save_pretrained('gpt2_')

from transformers import pipeline
fill_masker = pipeline(task ='text-generation', model="/content/gpt2_")

Heres how you implement greedy

fill_masker("hello my name is inigo montoya you",max_length=20)

maybe you can ask the huggingface community for alternative searching methods

---

Also here how you do beam search with huggingface

fill_masker("hello my name is inigo montoya you",num_return_sequences=20, num_beams=60,batch_size=20)

refer to this link to explore different searches https://huggingface.co/docs/transformers/main_classes/text_generation