Instruction Following Evaluation on AlpacaEval 2.0 (test)
54.17LC% over π0DAR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DARResponse Length=19632026.02 | 54.17 | 0.23 | — | |
| RLOOResponse Length=20762026.02 | 52.25 | 0.14 | — | |
| GRPOResponse Length=20382026.02 | 50.5 | 0.16 | — | |
| Iter-SFTResponse Length=20042026.02 | 49.8 | 0.17 | — | |
| OursBase Model=Llama-3.1-Tulu-3-8B, Data Selection=30% easiest prompts2025.10 | 38.89 | — | 45.78 | |
| Full DatasetBase Model=Llama-3.1-Tulu-3-8B, Data Selection=Full Dataset2025.10 | 34.37 | — | 36.46 | |
| RandomBase Model=Llama-3.1-Tulu-3-8B, Data Selection=30% random prompts2025.10 | 33.48 | — | 35.45 | |
| OursBase Model=Mistral-7B-Instruct-v0.2, Data Selection=30% easiest prompts2025.10 | 31.43 | — | 39.01 | |
| Full DatasetBase Model=Mistral-7B-Instruct-v0.2, Data Selection=Full Dataset2025.10 | 29.24 | — | 29.94 | |
| RandomBase Model=Mistral-7B-Instruct-v0.2, Data Selection=30% random prompts2025.10 | 27.88 | — | 38.01 |