mt5-base-finetuned-fa
| Property | Value |
|---|---|
| Base Model | google/mt5-base |
| Task | Text Summarization |
| Language | Farsi |
| Framework | PyTorch 1.11.0+cu113 |
| Hugging Face Link | Model Repository |
What is mt5-base-finetuned-fa?
mt5-base-finetuned-fa is a specialized text summarization model built upon Google's MT5-base architecture, specifically fine-tuned for Farsi language processing. The model demonstrates strong performance metrics, achieving a ROUGE-1 score of 33.7, ROUGE-2 of 21.28, and ROUGE-L of 31.69, along with a impressive BERTScore of 74.52.
Implementation Details
The model was trained using a carefully configured hyperparameter setup, including a learning rate of 0.0005, batch size of 32 (achieved through gradient accumulation), and Adam optimizer with betas=(0.9,0.999). The training process spanned 5 epochs with linear learning rate scheduling and 250 warmup steps.
- Training utilized gradient accumulation steps of 8
- Implemented label smoothing factor of 0.1
- Achieved consistent generation length of 19.0 tokens
- Progressive improvement in validation metrics across training epochs
Core Capabilities
- Specialized in Farsi text summarization
- Robust performance with ROUGE metrics
- Consistent output generation length
- Optimized for production deployment with PyTorch
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized fine-tuning on Farsi text summarization, achieving competitive ROUGE scores while maintaining consistent generation length. The careful hyperparameter optimization and training process demonstrate its reliability for production applications.
Q: What are the recommended use cases?
The model is best suited for Farsi text summarization tasks where consistent, high-quality summaries are required. Its strong BERTScore of 74.52 suggests good semantic understanding and generation capabilities.
