mT5_multilingual_XLSum-finetuned-fa
| Property | Value |
|---|---|
| Base Model | csebuetnlp/mT5_multilingual_XLSum |
| Fine-tuned On | pn_summary dataset (Persian) |
| Author | ahmeddbahaa |
| Model Hub | Hugging Face |
What is mT5_multilingual_XLSum-finetuned-fa?
This model is a specialized version of mT5 that has been fine-tuned specifically for Persian language text summarization. Built upon the multilingual XLSum foundation, it demonstrates impressive performance metrics including a ROUGE-1 score of 45.12 and a BERTScore of 79.54, making it particularly effective for Persian content summarization tasks.
Implementation Details
The model was trained using carefully selected hyperparameters, including a learning rate of 0.0005, Adam optimizer with betas=(0.9,0.999), and a linear learning rate scheduler with 250 warmup steps. The training process utilized a batch size of 16 and ran for 5 epochs with label smoothing factor of 0.1.
- Training batch size: 16 (2 base * 8 gradient accumulation steps)
- Evaluation batch size: 2
- Optimizer: Adam with epsilon=1e-08
- Learning rate scheduling: Linear with warmup
Core Capabilities
- ROUGE-1 Score: 45.12
- ROUGE-2 Score: 26.25
- ROUGE-L Score: 39.96
- BERTScore: 79.54
- Average Generation Length: 48.72 tokens
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in Persian text summarization, achieving strong performance metrics while leveraging the multilingual capabilities of mT5. Its high ROUGE and BERTScore metrics indicate reliable summarization quality for Persian content.
Q: What are the recommended use cases?
The model is best suited for Persian text summarization tasks, particularly where automatic generation of concise summaries is needed. It can be effectively used in news aggregation, content briefing, and document summarization applications focusing on Persian language content.
