opus-mt-en-mr
| Property | Value |
|---|---|
| License | apache-2.0 |
| Architecture | transformer-align |
| BLEU Score | 22.0 |
| chr-F Score | 0.397 |
What is opus-mt-en-mr?
opus-mt-en-mr is a specialized machine translation model developed by Helsinki-NLP for translating English text to Marathi. Built on the OPUS dataset, it employs a transformer-align architecture with normalization and SentencePiece pre-processing techniques. The model has demonstrated significant performance with a BLEU score of 22.0 on the Tatoeba test set.
Implementation Details
The model utilizes a transformer-based architecture optimized for English to Marathi translation. It implements both PyTorch and TensorFlow frameworks, making it versatile for different deployment environments.
- Pre-processing: Implements normalization and SentencePiece tokenization
- Framework compatibility: Supports both PyTorch and TensorFlow
- Evaluation metrics: BLEU score of 22.0 and chr-F score of 0.397
Core Capabilities
- Direct translation from English to Marathi
- Support for inference endpoints
- Optimized for production deployment
- Comprehensive evaluation on Tatoeba test set
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in English to Marathi translation using a transformer-align architecture, with specific optimization for the language pair and comprehensive evaluation metrics. The combination of normalization and SentencePiece pre-processing enhances its translation quality.
Q: What are the recommended use cases?
The model is ideal for applications requiring English to Marathi translation, such as content localization, document translation, and multilingual applications. With its inference endpoints support, it's suitable for both batch processing and real-time translation tasks.
