Instruction Following on AlpacaEval (test)
3,213Helpfulness ScoreFine-tuned
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Fine-tunedBackbone=LLaMA3-Inst-8B, CRqt=12025.04 | 3,213 | — | |
| IMPART-QTBackbone=LLaMA3-Inst-8B, CRqt=322025.04 | 2,827 | — | |
| DARE-QTBackbone=LLaMA3-Inst-8B, CRqt=322025.04 | 2,824 | — | |
| Delta-CoMeBackbone=LLaMA3-Inst-8B, CRqt=322025.04 | 2,664 | — | |
| BitDeltaBackbone=LLaMA3-Inst-8B, CRqt=322025.04 | 2,635 | — | |
| Fine-tunedBackbone=LLaMA2-Chat-13B, CRqt=12025.04 | 1,839 | — | |
| BitDeltaBackbone=LLaMA2-Chat-13B, CRqt=322025.04 | 1,830 | — | |
| DARE-QTBackbone=LLaMA2-Chat-13B, CRqt=322025.04 | 1,768 | — | |
| IMPART-QTBackbone=LLaMA2-Chat-13B, CRqt=322025.04 | 1,765 | — | |
| Fine-tunedBackbone=LLaMA2-Chat-7B, CRqt=12025.04 | 1,563 | — | |
| Delta-CoMeBackbone=LLaMA2-Chat-13B, CRqt=322025.04 | 1,552 | — | |
| IMPART-QTBackbone=LLaMA2-Chat-7B, CRqt=322025.04 | 1,255 | — | |
| BitDeltaBackbone=LLaMA2-Chat-7B, CRqt=322025.04 | 1,187 | — | |
| DARE-QTBackbone=LLaMA2-Chat-7B, CRqt=322025.04 | 1,176 | — | |
| Delta-CoMeBackbone=LLaMA2-Chat-7B, CRqt=322025.04 | 1,085 | — | |
| PEPOBase Model=Mistral-7B2026.02 | 87.2 | — | |
| DPOBase Model=Mistral-7B2026.02 | 85.3 | — | |
| χ2POBase Model=Mistral-7B2026.02 | 82.8 | — | |
| PEPOBase Model=Llama-3-8B2026.02 | 76.1 | — | |
| SFTBackbone=Llama3.1-8B2026.05 | 74.2 | 43.87 | |
| SFT+DPOBase Model=Mistral-7B2026.02 | 72.4 | — | |
| BackboneBackbone=LLaMA2-Chat-13B, CRqt=12025.04 | 71 | — | |
| χ2POBase Model=Llama-3-8B2026.02 | 70.7 | — | |
| SFT+DPOBase Model=Llama-3-8B2026.02 | 69.7 | — | |
| DPOBase Model=Llama-3-8B2026.02 | 69.3 | — | |
| VaccineBackbone=Llama3.1-8B2026.05 | 62.9 | 36.39 | |
| Abstract-CoT (Warm-up + RL)Model=Qwen3-8B, Tokens=2252026.04 | 60.8 | — | |
| Abstract-CoT (Warm-up + RL)Model=Qwen3-4B, Tokens=2132026.04 | 58.7 | — | |
| SFT + RLModel=Qwen3-8B, Tokens=4962026.04 | 58.4 | — | |
| LisaBackbone=Llama3.1-8B2026.05 | 58.2 | 37.93 | |
| SFT + RLModel=Qwen3-4B, Tokens=4672026.04 | 57.1 | — | |
| SFT (CoT)Model=Qwen3-8B, Tokens=4602026.04 | 57 | — | |
| Abstract-CoT (Warm-up)Model=Qwen3-8B, Tokens=2512026.04 | 55.9 | — | |
| Stepwise InternalizationModel=Qwen3-8B, Tokens=2452026.04 | 55.3 | — | |
| SFT (CoT)Model=Qwen3-4B, Tokens=4312026.04 | 55.2 | — | |
| SFT (no CoT)Model=Qwen3-8B, Tokens=2982026.04 | 54.9 | — | |
| Abstract-CoT (Warm-up)Model=Qwen3-4B, Tokens=2382026.04 | 54.3 | — | |
| BoosterBackbone=Llama3.1-8B2026.05 | 54.3 | 35.75 | |
| SFT (no CoT)Model=Qwen3-4B, Tokens=2762026.04 | 53.7 | — | |
| BaselineModel=Qwen3-8B, Tokens=3222026.04 | 52.4 | — | |
| Stepwise InternalizationModel=Qwen3-4B, Tokens=1742026.04 | 51.4 | — | |
| Abstract-CoT (RL-only)Model=Qwen3-8B, Tokens=2292026.04 | 50.4 | — | |
| BaselineModel=Qwen3-4B, Tokens=2982026.04 | 49.8 | — | |
| Abstract-CoT (RL-only)Model=Qwen3-4B, Tokens=2172026.04 | 47.8 | — | |
| Pause TokensModel=Qwen3-8B, Tokens=2122026.04 | 46.7 | — | |
| NA-SFTBackbone=Mistral-7B2024.05 | 43.2 | 54.33 | |
| Pause TokensModel=Qwen3-4B, Tokens=1982026.04 | 42.1 | — | |
| SFTBackbone=Mistral-7B2024.05 | 38.5 | 45.63 | |
| VlguardBackbone=Mistral-7B2024.05 | 37.5 | 43.27 | |
| BSOBackbone=Mistral-7B2024.05 | 37.4 | 41.75 | |
| Abstract-CoT (Warm-up + RL)Model=Granite 4.0 Micro (3B), Tokens=2192026.04 | 33.54 | — | |
| LisaBackbone=Mistral-7B2024.05 | 33.1 | 41.35 | |
| Vaccine-SFTBackbone=Mistral-7B2024.05 | 33 | 10.68 | |
| SFT + RLModel=Granite 4.0 Micro (3B), Tokens=4182026.04 | 31.97 | — | |
| BaselineModel=Granite 4.0 Micro (3B), Tokens=3412026.04 | 29.13 | — | |
| SFT (CoT)Model=Granite 4.0 Micro (3B), Tokens=4422026.04 | 29.06 | — | |
| Abstract-CoT (Warm-up)Model=Granite 4.0 Micro (3B), Tokens=2342026.04 | 28.69 | — | |
| SFT (no CoT)Model=Granite 4.0 Micro (3B), Tokens=3172026.04 | 28.42 | — | |
| Abstract-CoT (RL-only)Model=Granite 4.0 Micro (3B), Tokens=2512026.04 | 28.26 | — | |
| Stepwise InternalizationModel=Granite 4.0 Micro (3B), Tokens=1982026.04 | 27.95 | — | |
| Pause TokensModel=Granite 4.0 Micro (3B), Tokens=2232026.04 | 23.84 | — | |
| DeepAlignBackbone=Llama3.1-8B2026.05 | 23.7 | 33.64 | |
| BackboneBackbone=LLaMA2-Chat-7B, CRqt=12025.04 | 10 | — | |
| BackboneBackbone=LLaMA3-Inst-8B, CRqt=12025.04 | 8 | — | |
| SBRBackbone=Llama3.1-8B2026.05 | 6.2 | 45.82 |