sage-fredt5-distilled-95m
| Property | Value |
|---|---|
| Model Size | 95M parameters |
| Framework | PyTorch |
| License | MIT |
| File Size | 0.383 GB |
| Developer | SberDevices, AGI NLP |
What is sage-fredt5-distilled-95m?
sage-fredt5-distilled-95m is a specialized Russian language model designed for correcting spelling, punctuation, and case errors in text. It's a distilled version of the larger FRED-T5-1.7B architecture, optimized for efficiency while maintaining high accuracy. The model was trained on an extensive dataset derived from Russian Wikipedia and video transcripts, with artificially introduced errors using the SAGE library.
Implementation Details
The model leverages the T5 architecture and has been specifically optimized through distillation to achieve a compact 95M parameter size while maintaining strong performance. It can be easily implemented using the Hugging Face Transformers library and supports both CPU and GPU inference.
- Trained on synthetic error dataset created using SAGE library
- Supports batch processing and variable length inputs
- Optimized for Russian language text correction
- Maintains high accuracy across spelling, punctuation, and case corrections
Core Capabilities
- Spelling error correction with up to 83.5% precision on RUSpellRU dataset
- Punctuation correction achieving 86.8% precision
- Case correction with 94.4% precision
- Multi-domain support including medical texts and technical documentation
- Real-time text normalization capabilities
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its efficient size while maintaining competitive performance metrics against larger models like GPT-3.5 and GPT-4 in Russian text correction tasks. It's specifically optimized for Russian language and shows strong performance across various text domains.
Q: What are the recommended use cases?
The model is ideal for applications requiring Russian text normalization, including content management systems, text editors, document processing systems, and automated proofreading tools. It's particularly effective for handling informal text with typical spelling and punctuation errors.
