Relation Extraction on Literature Benchmarks Biographical and PG-Fiction
91.7Biographical F1SmolLM3-3B LitTune 0s
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SmolLM3-3B LitTune 0sBase Model=SmolLM3-3B, Tuning Regime=LitTune, Prompt Style=0-shot (0s)2026.06 | 91.7 | 74.9 | 83.3 | 65 | 99.9 | — | |
| Llama-3.2-3B MixTune 2sBase Model=Llama-3.2-3B, Tuning Regime=MixTune, Prompt Style=2-shot (2s)2026.06 | 91.3 | 73.7 | 82.5 | 62 | 99.9 | — | |
| GPT-5.4Prompt Style=0-shot2026.06 | 83.2 | 32.4 | 57.8 | 50 | 86.7 | — | |
| Claude Sonnet 4.6Prompt Style=0-shot2026.06 | 72.5 | 33.4 | 53 | 41 | 90.3 | — | |
| Best SLMPrompting=schema-enumerated, Selection=strongest tuned SLM per benchmark2026.06 | — | — | — | — | — | 91.7 | |
| ClaudeInference mode=zero-shot2026.06 | — | — | — | — | — | 72.5 | |
| GPT-5.4Inference mode=zero-shot2026.06 | — | — | — | — | — | 83.2 | |
| RoB-baseBackbone=RoBERTa, Parameters=125M, Inference mode=fine-tuned per benchmark2026.06 | — | — | — | — | — | 90.6 |