Question Answering on MedQA, PubMedQA, MedMCQA, and HeadQA Aggregate
50.63Average AccuracyOLMo-2-7B-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| OLMo-2-7B-InstructBackbone=OLMo-2-7B-Instruct2026.04 | 50.63 | |
| DPOBackbone=Llama-3.2-3B-Instruct2026.04 | 50.09 | |
| DPOBackbone=OLMo-2-7B-Instruct2026.04 | 48.8 | |
| SKUBackbone=OLMo-2-7B-Instruct2026.04 | 48.17 | |
| RetainOnlyFTBackbone=OLMo-2-7B-Instruct2026.04 | 48 | |
| EUBackbone=OLMo-2-7B-Instruct, Lambda (λ)=0.62026.04 | 47.53 | |
| UnlearningBackbone=Llama-3.2-3B-Instruct, Lambda (λ)=0.62026.04 | 46.95 | |
| UnlearningBackbone=OLMo-2-7B-Instruct, Lambda (λ)=0.62026.04 | 46.33 | |
| EraserBackbone=OLMo-2-7B-Instruct, Lambda (λ)=0.62026.04 | 45.82 | |
| EraserBackbone=Llama-3.2-3B-Instruct, Lambda (λ)=0.62026.04 | 45 | |
| RetainOnlyFTBackbone=Llama-3.2-3B-Instruct2026.04 | 44.96 | |
| EUBackbone=Llama-3.2-3B-Instruct, Lambda (λ)=0.62026.04 | 44.68 | |
| Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct2026.04 | 42.9 | |
| SKUBackbone=Llama-3.2-3B-Instruct2026.04 | 41.98 | |
| UnlearningBackbone=Llama-3.2-1B-Instruct, Lambda (λ)=0.62026.04 | 38.98 | |
| RetainOnlyFTBackbone=Llama-3.2-1B-Instruct2026.04 | 38.2 | |
| EraserBackbone=OLMo-2-1B-Instruct, Lambda (λ)=0.952026.04 | 37.64 | |
| EraserBackbone=Llama-3.2-1B-Instruct, Lambda (λ)=0.62026.04 | 37.2 | |
| SKUBackbone=OLMo-2-1B-Instruct2026.04 | 37.13 | |
| DPOBackbone=Llama-3.2-1B-Instruct2026.04 | 37.12 | |
| Llama-3.2-1B-InstructBackbone=Llama-3.2-1B-Instruct2026.04 | 36.72 | |
| EUBackbone=Llama-3.2-1B-Instruct, Lambda (λ)=0.62026.04 | 36.05 | |
| RetainOnlyFTBackbone=OLMo-2-1B-Instruct2026.04 | 35.67 | |
| UnlearningBackbone=OLMo-2-1B-Instruct, Lambda (λ)=0.952026.04 | 35.59 | |
| EUBackbone=OLMo-2-1B-Instruct, Lambda (λ)=0.952026.04 | 35.41 | |
| DPOBackbone=OLMo-2-1B-Instruct2026.04 | 35.07 | |
| OLMo-2-1B-InstructBackbone=OLMo-2-1B-Instruct2026.04 | 33.86 | |
| SKUBackbone=Llama-3.2-1B-Instruct2026.04 | 33.14 |