ij5/whitespace-correction

Basic Inference

from transformers import T5TokenizerFast, T5ForConditionalGeneration

tokenizer = T5TokenizerFast.from_pretrained('ij5/whitespace-correction')
model = T5ForConditionalGeneration.from_pretrained('ij5/whitespace-correction')

def fix_whitespace(text):
    inputs = f"띄어쓰기 교정: {text}"
    tokenized = tokenizer(inputs, max_length=128, truncation=True, return_tensors='pt').to('cuda')
    output_ids = model.generate(
        input_ids=tokenized['input_ids'],
        attention_mask=tokenized['attention_mask'],
        max_length=128,
    )
    return tokenizer.decode(output_ids[0], skip_special_tokens=True)


print(fix_whitespace("흔들 리는 가지 사이로 불쑥 바람의 형상 이 드 러나기라도 할 것처럼."))
# result: 흔들리는 가지 사이로 불쑥 바람의 형상이 드러나기라도 할 것처럼.

correction

This model is a fine-tuned version of paust/pko-t5-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0160

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.0243	1.0	1688	0.0183
0.0172	2.0	3376	0.0165
0.0126	3.0	5064	0.0160

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0