Formulation QA on Formulation QA (OOD)
39.4AccuracyOriginal
Evaluation Results
| Method | Links | |
|---|---|---|
| OriginalToken=0, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 39.4 | |
| w GPT-5Token=50*3, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 38.3 | |
| SciDCToken=30*5, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 38.3 | |
| w wikiToken=800*5, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 36.1 | |
| vanilla RAGToken=512*3, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 35.6 | |
| w summaryToken=30*5, Backbone=LLaMA-3.2-3B-Instruct2026.04 | 33.33 |