| license: mit | |
| tags: | |
| - deberta | |
| - text-complexity | |
| - onnx | |
| - regression | |
| library_name: transformers | |
| # MoA Complexity Regressor v2 | |
| DeBERTa-v3-small fine-tuned for text complexity scoring (regression). | |
| ## Metrics | |
| - **MAE**: 0.0418 | |
| - **R2**: 0.954 | |
| - **Effort Accuracy**: 85.5% | |
| ## Effort Tiers | |
| | Score Range | Label | | |
| |------------|-------| | |
| | 0.00-0.08 | trivial | | |
| | 0.08-0.18 | light | | |
| | 0.18-0.32 | moderate | | |
| | 0.32-0.52 | heavy | | |
| | 0.52-0.72 | intensive | | |
| | 0.72-1.00 | extreme | | |
| ## Usage | |
| Load the ONNX model with onnxruntime, tokenize input with DeBERTa-v3-small tokenizer. | |