Single-label multi-class topic classification on SCOTUS (test)
77.5Micro-F1CPE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CPEtraining_setting=end-to-end fine-tuning, backbone=LegalBERT2025.12 | 77.5 | 67.3 | |
| LegalLongformertraining_setting=end-to-end fine-tuning2025.12 | 76.6 | 66.9 | |
| Hi-LegalBERTtraining_setting=end-to-end fine-tuning2025.12 | 76.4 | 66.5 | |
| LSGtraining_setting=end-to-end fine-tuning2025.12 | 73.3 | 63.7 | |
| LexLM-LargeFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 71.1 | 43.9 | |
| HATtraining_setting=end-to-end fine-tuning2025.12 | 70.5 | 59.1 | |
| LegalModernBERT-largeModel size=large, Pre-training strategy=domain-adapted (continued pre-training)2026.06 | 69.3 | 78.5 | |
| ROBERTa-LargeFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 68.9 | 41.4 | |
| ModernBERT-largeModel size=large, Pre-training strategy=vanilla2026.06 | 68.6 | 77.9 | |
| LegalModernBERT-baseModel size=base, Pre-training strategy=domain-adapted (continued pre-training)2026.06 | 67.5 | 77.5 | |
| LexLM-BaseFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 66.9 | 37.7 | |
| POL-BERTFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 66.3 | 39.5 | |
| RoBERTa-largeModel size=large2026.06 | 66.3 | 75.5 | |
| Best Reported in [6]Model size=base2026.06 | 65.9 | 76.6 | |
| ModernBERT-baseModel size=base, Pre-training strategy=vanilla2026.06 | 65.3 | 75.7 | |
| ROBERTa-BaseFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 65 | 36 | |
| LegalModernBERT-base from scratchModel size=base, Pre-training strategy=from scratch2026.06 | 63.4 | 75.2 | |
| LegalModernBERT-large from scratchModel size=large, Pre-training strategy=from scratch2026.06 | 63.4 | 75.2 | |
| CL-BERTFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 62.9 | 33.8 | |
| LegalBERTFine-tuning epochs=1, Architecture=Warm-start Longformer, Tokens=20482023.05 | 60.9 | 31.2 |