Instruction following on AlpacaEval High-Variance (Top 20%) 2.0
11.6Reward ScorerDPO (Best-of-K)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| rDPO (Best-of-K)Generator=Llama-3.1-8B-Instruct, Type=Training2026.03 | 11.6 | 9.05 | -6.5 | 347 | |
| rDPO + DARC-ϵGenerator=Llama-3.1-8B-Instruct, Type=Training2026.03 | 11.56 | 8.65 | -5.74 | 329 | |
| cDPO (Best-of-K)Generator=Llama-3.1-8B-Instruct, Type=Training2026.03 | 8.88 | 9.4 | -9.92 | 328 | |
| cDPO + DARC-ϵGenerator=Llama-3.1-8B-Instruct, Type=Training2026.03 | 8.81 | 9.21 | -9.61 | 328 | |
| DeALGenerator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.89 | 10.15 | -12.41 | 371 | |
| rDPO (Best-of-K)Generator=Qwen2.5-7B-Instruct, Type=Training2026.03 | 7.8 | 7.72 | -7.64 | 172 | |
| rDPO + DARC-ϵGenerator=Qwen2.5-7B-Instruct, Type=Training2026.03 | 7.7 | 7.28 | -6.86 | 164 | |
| Base (Best-of-K)Generator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.48 | 10.22 | -12.96 | 381 | |
| BoP(HedgeTune)Generator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.28 | 8.12 | -8.96 | 381 | |
| DARC-ϵ(ours)Generator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.26 | 8.1 | -8.94 | 381 | |
| DARC(ours)Generator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.24 | 8.1 | -8.96 | 381 | |
| RBoNGenerator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.22 | 8.4 | -9.58 | 376 | |
| DARC-τ(ours)Generator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.21 | 8.55 | -9.89 | 381 | |
| MC-DropoutGenerator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.19 | 8.15 | -9.11 | 380 | |
| CautionGenerator=Llama-3.1-8B-Instruct, Type=Inference2026.03 | 7.17 | 8.34 | -9.51 | 381 | |
| cDPO (Best-of-K)Generator=Qwen2.5-7B-Instruct, Type=Training2026.03 | 7.17 | 7.3 | -7.43 | 196 | |
| cDPO + DARC-ϵGenerator=Qwen2.5-7B-Instruct, Type=Training2026.03 | 7.12 | 7.29 | -7.46 | 196 | |
| DeALGenerator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 3.45 | 6.89 | -10.33 | 256 | |
| Base (Best-of-K)Generator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.79 | 5.46 | -8.13 | 263 | |
| DARC-ϵ(ours)Generator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.59 | 3.76 | -4.93 | 265 | |
| BoP(HedgeTune)Generator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.53 | 4.09 | -5.65 | 265 | |
| MC-DropoutGenerator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.35 | 4 | -5.65 | 262 | |
| DARC(ours)Generator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.34 | 4.42 | -6.5 | 263 | |
| DARC-τ(ours)Generator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.31 | 3.84 | -5.37 | 249 | |
| CautionGenerator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.3 | 4.07 | -5.84 | 264 | |
| RBoNGenerator=Qwen2.5-7B-Instruct, Type=Inference2026.03 | 2.25 | 4.15 | -6.05 | 258 |