mbart-large-50-finetuned-opus-en-pt-translation
| Property | Value |
|---|---|
| Author | Narrativa |
| Paper | Original mBART-50 Paper |
| Dataset | OPUS100 & OpusBook |
| BLEU Score | 20.61 |
What is mbart-large-50-finetuned-opus-en-pt-translation?
This is a specialized neural machine translation model based on mBART-50 architecture, fine-tuned specifically for English to Portuguese translation. It leverages the powerful multilingual capabilities of mBART-50, which supports 50 languages, and has been optimized using the OPUS100 dataset to deliver high-quality translations between English and Portuguese.
Implementation Details
The model utilizes the multilingual denoising pretraining approach, where it learns from corrupted text input to reconstruct the original content. The pretraining process involves masking 35% of words in each instance, with span lengths following a Poisson distribution (λ = 3.5). The model employs language ID symbols (LID) to manage different language pairs effectively.
- Built on mBART-50 architecture supporting 50 languages
- Fine-tuned on OPUS100 dataset with up to 1M sentence pairs
- Uses specialized tokenization with MBart50TokenizerFast
- Implements forced decoding with language-specific tokens
Core Capabilities
- High-quality English to Portuguese translation
- Handles complex sentence structures and maintains context
- Supports batch processing for efficient translation
- Achieves a BLEU score of 20.61 on test sets
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its specialized fine-tuning on the OPUS100 dataset specifically for English-Portuguese translation, while maintaining the robust multilingual foundation of mBART-50. Its architecture ensures high-quality translations while preventing cross-lingual sentence overlap in training data.
Q: What are the recommended use cases?
The model is ideal for professional translation services, content localization, and applications requiring accurate English to Portuguese translation. It's particularly suitable for scenarios where maintaining context and handling complex sentence structures is crucial.
