| Property | Value |
|---|---|
| Parameter Count | 140M |
| Model Type | Encoder-Decoder |
| Base Architecture | BERT (dbmdz/bert-base-turkish-cased) |
| Training Dataset | MLSUM Turkish |
| Performance | Rouge2 F-measure: 29.48% |
What is bert2bert_shared-turkish-summarization?
This is a specialized Turkish language model designed for text summarization tasks, built upon the BERT architecture with shared encoder-decoder components. Developed by Manuel Romero, it leverages the dbmdz/bert-base-turkish-cased checkpoint and has been fine-tuned on the MLSUM Turkish dataset, which is part of a larger multilingual summarization corpus.
Implementation Details
The model utilizes a BERT2BERT architecture with 140M parameters, implementing both encoding and decoding capabilities for Turkish text summarization. It employs PyTorch framework and supports inference endpoints for production deployment.
- Achieves 32.41% Rouge2 precision and 28.65% Rouge2 recall on test sets
- Supports maximum input length of 512 tokens
- Implements efficient text generation with attention masking
- Uses safetensors for model weight storage
Core Capabilities
- Automatic summarization of Turkish news articles
- Handle long-form text input with truncation
- Generate concise, coherent summaries
- Support for batch processing and GPU acceleration
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically optimized for Turkish language summarization, using a shared BERT architecture that reduces model size while maintaining performance. It's one of the few specialized Turkish summarization models available with documented performance metrics.
Q: What are the recommended use cases?
The model is particularly well-suited for summarizing Turkish news articles, research papers, and long-form content. It's ideal for applications requiring automated content summarization in Turkish media monitoring, content aggregation, and information extraction systems.