serdarcaglar/roberta-base-turkish-scientific-cased-ONNX

Model Card: Turkish Scientific RoBERTa ONNX

Model Description

ONNX version of roberta-base-turkish-scientific-cased, specialized for Turkish scientific text analysis.

Intended Use

• Scientific text analysis in Turkish
• Text comprehension
• Fill-mask predictions
• Scientific text summarization

Training Data

• Source: Turkish scientific article abstracts from trdizin, yöktez, and t.k.
• Training Duration: 3+ days
• Steps: 2M
• Built from scratch, no fine-tuning

Technical Specifications

• Base Architecture: RoBERTa
• Tokenizer: BPE (Byte Pair Encoding)
• Format: ONNX
• Original Model: serdarcaglar/roberta-base-turkish-scientific-cased

Performance and Limitations

• Optimized for scientific domain in Turkish
• Not tested for general domain text
• ONNX format optimized for inference

Requirements

• onnxruntime
• transformers
• torch

License and Usage

• Follow original model license
• Users responsible for compliance

Citation

@misc{caglar2024roberta,
  author = {Çağlar, Serdar},
  title = {Roberta-base-turkish-scientific-cased},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/serdarcaglar/roberta-base-turkish-scientific-cased}
}

Contact

Serdar ÇAĞLAR ([email protected])