gemma-2b-it-summarization-paper

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

outputs

This model is a fine-tuned version of google/gemma-2b-it on AI hub 논문자료 요약 dataset.(https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=&topMenu=&aihubDataSe=data&dataSetSn=90)

Model Developers: Eonseon Park

Model description

Based on Gemma-2b-it, the model was developed through customized fine-tuning for summarizing tasks of different paper data. Based on existing natural language processing (NLP) capabilities, it is designed to effectively summarize the paper text. The model can learn a large-scale paper dataset to deliver the core content of the paper concisely and accurately, and can be applied to papers in various disciplines.

Intended uses & limitations

Intended uses : Paper summary, information retrieval and document processing, automated reviews

limitations : Creative Summary, Model Size and Performance

Training and evaluation data

training_논문 and validation_논문 in AI hub 논문자료 요약 dataset

Training procedure

After loading the json file, the dataset was organized in a chat format to extract only the 'original_text' and 'summary_text' required for learning and enter them into the model. It take 17 hours for training. train code : https://colab.research.google.com/drive/1z8ER-AfVcccDXFWsRzuD-m2LxTAjuaTR

How to use

messages = [
    {
        "role": "user",
        "content": "다음 글을 요약해주세요:\n\n{}".format(doc)
    }
]
prompt = pipe_finetuned.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

outputs = pipe_finetuned(
    prompt,
    do_sample=True,
    temperature=0.2,
    top_k=50,
    top_p=0.95,
    add_special_tokens=True
)
print(outputs[0]["generated_text"][len(prompt):])

inference code : https://colab.research.google.com/drive/1XzwA1fbfc3QttLHCrhVHBPDZIiL2a_wB

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2
• mixed_precision_training: Native AMP

Framework versions

• PEFT 0.13.0
• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0