Instruction Following on IFEval loose
78.1Avg@8Base
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BaseBackbone=Llama3.1-8B-Instruct2025.08 | 78.1 | — | |
| BaseBackbone=Qwen2.5-7B-Instruct2025.08 | 73.94 | — | |
| Qwen2.5-7B-InstructTraining Pipeline=PSFT → GRPO2025.08 | 73.73 | — | |
| PSFTBackbone=Qwen2.5-7B-Instruct2025.08 | 73.03 | — | |
| Qwen2.5-7B-InstructTraining Pipeline=PSFT2025.08 | 73.03 | — | |
| Llama3.1-8B-InstructTraining Pipeline=PSFT → GRPO2025.08 | 71.02 | — | |
| PSFTBackbone=Llama3.1-8B-Instruct2025.08 | 69.75 | — | |
| Llama3.1-8B-InstructTraining Pipeline=PSFT2025.08 | 69.75 | — | |
| SFT-KLBackbone=Qwen2.5-7B-Instruct2025.08 | 55.44 | — | |
| PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.08 | 55.07 | — | |
| SFTBackbone=Qwen2.5-7B-Instruct2025.08 | 54.42 | — | |
| Qwen2.5-7B-InstructTraining Pipeline=SFT2025.08 | 54.42 | — | |
| Qwen2.5-7B-InstructTraining Pipeline=SFT → GRPO2025.08 | 53.89 | — | |
| SFT-KLBackbone=Llama3.1-8B-Instruct2025.08 | 34.19 | — | |
| SFTBackbone=Llama3.1-8B-Instruct2025.08 | 33.45 | — | |
| Llama3.1-8B-InstructTraining Pipeline=SFT2025.08 | 33.45 | — | |
| PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.08 | 33.08 | — | |
| Llama3.1-8B-InstructTraining Pipeline=SFT → GRPO2025.08 | 31.85 | — | |
| LLaMACheckpoint=SFTMaxOOD2025.09 | — | 27 | |
| LLaMACheckpoint=SFTEnd2025.09 | — | 27 | |
| LLaMACheckpoint=RLEnd2025.09 | — | 27 | |
| QwenCheckpoint=SFTMaxOOD2025.09 | — | 63 | |
| QwenCheckpoint=SFTEnd2025.09 | — | 67 | |
| QwenCheckpoint=RLEnd2025.09 | — | 70 |