girt-t5-base

GIRT-Model

paper: https://arxiv.org/abs/2402.02632

demo: https://huggingface.co/spaces/nafisehNik/girt-space

This model is fine-tuned to generate issue report templates based on the input instruction provided. It has been fine-tuned on GIRT-Instruct data.

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# load model and tokenizer
model = AutoModelForSeq2SeqLM.from_pretrained('nafisehNik/girt-t5-base')
tokenizer = AutoTokenizer.from_pretrained(nafisehNik/girt-t5-base)

# method for computing issue report template generation
def compute(sample, top_p, top_k, do_sample, max_length, min_length):

    inputs = tokenizer(sample, return_tensors="pt").to('cpu')

    outputs = model.generate(
        **inputs,
        min_length= min_length,
        max_length=max_length,
        do_sample=do_sample,
        top_p=top_p,
        top_k=top_k).to('cpu')

    generated_texts = tokenizer.batch_decode(outputs, skip_special_tokens=False)
    generated_text = generated_texts[0]
    
    replace_dict = {
        '\n ': '\n',
        '</s>': '',
        '<pad> ': '',
        '<pad>': '',
        '<unk>!--': '<!--',
        '<unk>': '',
    }
    
    postprocess_text = generated_text
    for key, value in replace_dict.items():
        postprocess_text = postprocess_text.replace(key, value)

    return postprocess_text

prompt = "YOUR INPUT INSTRUCTION"
result = compute(prompt, top_p = 0.92, top_k=0, do_sample=True, max_length=300, min_length=30)

Citation

@article{nikeghbal2024girt,
  title={GIRT-Model: Automated Generation of Issue Report Templates},
  author={Nikeghbal, Nafiseh and Kargaran, Amir Hossein and Heydarnoori, Abbas},
  journal={arXiv preprint arXiv:2402.02632},
  year={2024}
}