Relation Extraction on TACRED full (test)
71.1Micro F1 (Positive Class)Best SLM
Evaluation results
| Method | Links | |
|---|---|---|
| Best SLMDescription=strongest tuned SLM per benchmark, Prompting=schema-enumerated, Selection=maximum over 30 configurations with matched shots2026.06 | 71.1 | |
| RoB-largeBackbone=RoBERTa, Parameters=355M, Inference mode=fine-tuned per benchmark, Seeding=single-seed2026.06 | 65.3 | |
| RoB-baseBackbone=RoBERTa, Parameters=125M, Inference mode=fine-tuned per benchmark, Seeding=single-seed2026.06 | 64.5 | |
| GPT-5.4Type=frontier model, Inference mode=zero-shot2026.06 | 53.1 | |
| ClaudeType=frontier model, Inference mode=zero-shot2026.06 | 44.5 |