<html> <head> <title>Model Card - WangchanBERTa QA Fine-tuned</title> </head> <body> <h1>Model Card</h1> <p><strong>Shared by :</strong> Sirinoot Ketkham</p> <p><strong>Model Name :</strong> WangchanBERTa QA Fine-tuned</p> <p><strong>Model ID:</strong> wangchanberta-qa-finetuned</p> <p><strong>Model type:</strong> chatbot for question-answering</p> <p><strong>Language(s) (NLP):</strong> th</p> <p><strong>Finetuned from model :</strong> wangchanberta-base-att-spm-uncased.</p> <h2>Model Details</h2> <h3>Model Description</h3> <p>The WangchanBERTa QA Fine-tuned model is a Thai question-answering model that has been fine-tuned from the base model, WangchanBERTa, using the Thai QA corpus dataset. The model has been trained to understand and answer questions in Thai based on the given context. It is suitable for various question-answering tasks in the Thai language.</p> <p>The model is based on the transformer architecture and has been fine-tuned using a training regime with a batch size of 32 and trained for 7 epochs. The training process utilized the Thai QA corpus dataset, which consists of approximately 17,000 question-answer pairs.nswer pairs. for study and research purposes only.</p> <h3>Speeds, Times</h3> <p>Due to the limited size of the training dataset, consisting of only 17,000 question-answer pairs, and the relatively low number of training epochs set to 7, the training process is completed within a relatively short time frame of approximately 13 minutes.</p> <p>It is essential to review and thoroughly understand the model's outputs before relying on them for critical or sensitive tasks. Regular monitoring and evaluation of the model's performance and ethical implications are recommended.</p> <h2>Evaluation</h2> <h3>Metrics</h3> <table> <tr> <th>Metric</th> <th>Score</th> </tr> <tr> <td>Accuracy</td> <td>0.924</td> </tr> <tr> <td>Precision</td> <td>0.958</td> </tr> <tr> <td>Recall</td> <td>0.954</td> </tr> <tr> <td>F1 Score</td> <td>0.914</td> </tr> <tr> <td>ROUGE-1 Score</td> <td>0.928</td> </tr> <tr> <td>ROUGE-2 Score</td> <td>0.529</td> </tr> <tr> <td>ROUGE-L Score</td> <td>0.928</td> </tr> <tr> <td>BLEURT Score</td> <td>0.830</td> </tr> <tr> <td>BLEU Score</td> <td>0.917</td> </tr> </table> <h2>Model Examination</h2> <ul> <li> <strong>Context:</strong> ประเทศเวเนซุเอลาตั้งอยู่ในทวีปใด อเมริกาใต้<br> <strong>Question:</strong> ประเทศเวเนซุเอลาตั้งอยู่ในทวีปใด </li> <li> <strong>Context:</strong> บิดาของหม่อมงามจิตต์ บุรฉัตร ณ อยุธยา คือใคร พันโท พระสารสาสน์พลขันธ์<br> <strong>Question:</strong> บิดาของหม่อมงามจิตต์ บุรฉัตร ณ อยุธยา คือใคร </li> <li> <strong>Context:</strong> นวนิยายแฟนตาซีเกี่ยวกับแวมไพร์ชื่อ แรกรัตติกาล หรือ Twilight ประพันธ์โดยใคร สเตเฟนี เมเยอร์<br> <strong>Question:</strong> นวนิยายแฟนตาซีเกี่ยวกับแวมไพร์ชื่อ แรกรัตติกาล หรือ Twilight ประพันธ์โดยใคร </li> </ul> <h2>References</h2> <p>Please note that the base model, WangchanBERTa, was developed by the AI Research (AIR) team at Thai NLP and can be accessed at the following GitHub repository: <a href="https://github.com/vistec-AI/thai2transformers">https://github.com/vistec-AI/thai2transformers</a>.</p> <p>For further details, documentation, and updates, please refer to the model's GitHub repository and the Thai NLP community resources.</p> </body> </html>