README.md

metadata

license: mit

How to use

from transformers import MT5Tokenizer, MT5ForConditionalGeneration

tokenizer = MT5Tokenizer.from_pretrained('/Users/siwaboon/Desktop/Artifacts/thaisum/checkpoint-21000')
model = MT5ForConditionalGeneration.from_pretrained('/Users/siwaboon/Desktop/Artifacts/thaisum/checkpoint-21000')

text = "some news with head line"

tokenized_text = tokenizer(text, truncation=True, padding=True, return_tensors='pt')
    
source_ids = tokenized_text['input_ids'].to("cpu", dtype = torch.long)
source_mask = tokenized_text['attention_mask'].to("cpu", dtype = torch.long)
    
generated_ids = model.generate(
    input_ids = source_ids,
    attention_mask = source_mask, 
    max_length=512,
    num_beams=5,
    repetition_penalty=1, 
    length_penalty=1, 
    early_stopping=True,
    no_repeat_ngram_size=2
)

pred = tokenizer.decode(generated_ids[0], skip_special_tokens=True, clean_up_tokenization_spaces=True)