| Property | Value |
|---|---|
| Developer | Google Research (Elron Bandel, Thibault Sellam, et al.) |
| Model Type | Text Classification |
| Base Architecture | BERT |
| Paper | BLEURT: Learning Robust Metrics for Text Generation |
| Downloads | 61,161 |
What is bleurt-tiny-512?
BLEURT-tiny-512 is a compact PyTorch implementation of Google's BLEURT (Bilingual Evaluation Understudy with Representations from Transformers) metric. This model is specifically designed for evaluating natural language generation quality, offering a more efficient alternative to the original implementation while maintaining robust performance.
Implementation Details
The model is built on BERT architecture and trained on WMT Metrics Shared Task data from 2017-2019, encompassing thousands of sentence pairs with human ratings from the news domain. The training dataset includes 5,360, 9,492, and 147,691 records for each respective year.
- PyTorch-based implementation for better compatibility and deployment
- Optimized for 512 token sequence length
- Trained on extensive WMT dataset with human annotations
- Supports direct text comparison and evaluation
Core Capabilities
- Text quality assessment and comparison
- Generation evaluation for machine translation
- Semantic similarity scoring
- Automated evaluation of natural language generation
Frequently Asked Questions
Q: What makes this model unique?
BLEURT-tiny-512 stands out for its efficient implementation of the BLEURT metric in PyTorch, making it more accessible for production environments while maintaining the robust evaluation capabilities of the original model.
Q: What are the recommended use cases?
The model is ideal for evaluating machine translation output, assessing natural language generation quality, and comparing generated text against reference sentences in production environments.