| Property | Value |
|---|---|
| Parameter Count | 244M |
| License | MIT |
| Architecture | T5-based |
| Training Data | 4 datasets (IlyaGusev/gazeta, xlsum, mlsum, wiki_lingua) |
What is rut5-base-absum?
rut5-base-absum is a specialized Russian language model designed for abstractive text summarization. Built on the foundation of rut5-base-multitask, this model has been fine-tuned on four diverse datasets to provide high-quality Russian text summaries. It features flexible summarization options, allowing users to specify either the desired word count or compression ratio.
Implementation Details
The model implements a T5 architecture with 244M parameters and uses PyTorch for computation. It accepts text input with optional prefixes to control summary length and employs beam search with configurable parameters for generation quality control.
- Supports both fixed word count and compression ratio-based summarization
- Uses beam search with customizable number of beams (default: 3)
- Implements repetition penalty for better summary quality
- Runs on CUDA-enabled devices for faster inference
Core Capabilities
- Abstractive summarization of Russian texts
- Flexible summary length control
- High-quality compression while maintaining coherence
- Support for long-form text inputs
Frequently Asked Questions
Q: What makes this model unique?
The model's ability to handle Russian text summarization with controllable output length and its training on multiple diverse datasets makes it particularly versatile for Russian language applications.
Q: What are the recommended use cases?
The model is ideal for automated news summarization, content brief generation, and document summarization tasks where Russian language support is required. It's particularly useful when specific summary lengths or compression ratios need to be maintained.