Multi-turn Instruction Following on MT Bench
9.16MT-Bench Score (GPT-4)Llama-3.1-Nemotron-70B-Instruct
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Llama-3.1-Nemotron-70B-InstructFeedback + Edit Protocol=true2025.03 | 9.16 | 2,614.4 | — | — | — | — | — | — | — | |
| Llama-3.3-70B-InstructFeedback + Edit Protocol=true2025.03 | 9.07 | 2,362.1 | — | — | — | — | — | — | — | |
| Llama-3.1-Nemotron-70B-InstructFeedback + Edit Protocol=false2025.03 | 8.98 | 2,199.8 | — | — | — | — | — | — | — | |
| Claude-3-5-Sonnet-202406202025.03 | 8.81 | 1,619.9 | — | — | — | — | — | — | — | |
| GPT-4o-2024-05-132025.03 | 8.74 | 1,752.2 | — | — | — | — | — | — | — | |
| Llama-3.1-405B-Instruct2025.03 | 8.49 | 1,664.7 | — | — | — | — | — | — | — | |
| IPOBackbone=Llama3-8B-Instruct2026.04 | 8.3 | — | — | — | — | — | — | — | — | |
| RMiPOBackbone=Llama3-8B-Instruct2026.04 | 8.3 | — | — | — | — | — | — | — | — | |
| Llama-3.3-70B-InstructFeedback + Edit Protocol=false2025.03 | 8.29 | 1,827.6 | — | — | — | — | — | — | — | |
| Llama-3.1-70B-Instruct2025.03 | 8.22 | 1,728.6 | — | — | — | — | — | — | — | |
| KTOBackbone=Llama3-8B-Instruct2026.04 | 8.2 | — | — | — | — | — | — | — | — | |
| SimPERBackbone=Llama3-8B-Instruct2026.04 | 8.2 | — | — | — | — | — | — | — | — | |
| InitialBackbone=Llama3-8B-Instruct2026.04 | 8.1 | — | — | — | — | — | — | — | — | |
| SLiCBackbone=Llama3-8B-Instruct2026.04 | 8.1 | — | — | — | — | — | — | — | — | |
| α-DPOBackbone=Llama3-8B-Instruct2026.04 | 8.1 | — | — | — | — | — | — | — | — | |
| DPOBackbone=Llama3-8B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| CPOBackbone=Llama3-8B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| SimPOBackbone=Llama3-8B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| α-DPOBackbone=Mistral-7B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| β-DPOBackbone=Llama3-8B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| ϵ-DPOBackbone=Llama3-8B-Instruct2026.04 | 8 | — | — | — | — | — | — | — | — | |
| SLiCBackbone=Mistral-7B-Instruct2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| IPOBackbone=Mistral-7B-Instruct2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| KTOBackbone=Llama3-8B-Base2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| ϵ-DPOBackbone=Mistral-7B-Instruct2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| SimPERBackbone=Mistral-7B-Instruct2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| RMiPOBackbone=Mistral-7B-Instruct2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| RMiPOBackbone=Llama3-8B-Base2026.04 | 7.8 | — | — | — | — | — | — | — | — | |
| DPOBackbone=Llama3-8B-Base2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| KTOBackbone=Mistral-7B-Instruct2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| SimPOBackbone=Llama3-8B-Base2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| α-DPOBackbone=Llama3-8B-Base2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| β-DPOBackbone=Mistral-7B-Instruct2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| β-DPOBackbone=Llama3-8B-Base2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| SimPERBackbone=Llama3-8B-Base2026.04 | 7.7 | — | — | — | — | — | — | — | — | |
| DPOBackbone=Mistral-7B-Instruct2026.04 | 7.6 | — | — | — | — | — | — | — | — | |
| SLiCBackbone=Llama3-8B-Base2026.04 | 7.6 | — | — | — | — | — | — | — | — | |
| SimPOBackbone=Mistral-7B-Instruct2026.04 | 7.6 | — | — | — | — | — | — | — | — | |
| ϵ-DPOBackbone=Llama3-8B-Base2026.04 | 7.6 | — | — | — | — | — | — | — | — | |
| InitialBackbone=Mistral-7B-Instruct2026.04 | 7.5 | — | — | — | — | — | — | — | — | |
| CPOBackbone=Mistral-7B-Instruct2026.04 | 7.5 | — | — | — | — | — | — | — | — | |
| α-DPOBackbone=Mistral-7B-Base2026.04 | 7.5 | — | — | — | — | — | — | — | — | |
| SimPERBackbone=Mistral-7B-Base2026.04 | 7.5 | — | — | — | — | — | — | — | — | |
| RMiPOBackbone=Mistral-7B-Base2026.04 | 7.5 | — | — | — | — | — | — | — | — | |
| InstructModel=Qwen3-4B2026.05 | 7.45 | — | — | — | — | — | — | — | — | |
| GW-DPOvariant=bi.2026.06 | 7.425 | — | — | — | — | — | 8.3 | 6.55 | 1.75 | |
| baselinetraining_config=baseline2026.06 | 7.406 | — | — | — | — | — | 8.025 | 6.788 | 1.238 | |
| SLiCBackbone=Mistral-7B-Base2026.04 | 7.4 | — | — | — | — | — | — | — | — | |
| IPOBackbone=Llama3-8B-Base2026.04 | 7.4 | — | — | — | — | — | — | — | — | |
| CPOBackbone=Llama3-8B-Base2026.04 | 7.4 | — | — | — | — | — | — | — | — | |
| ϵ-DPOBackbone=Mistral-7B-Base2026.04 | 7.4 | — | — | — | — | — | — | — | — | |
| DPOtraining_config=DPO2026.06 | 7.363 | — | — | — | — | — | 8.1 | 6.625 | 1.475 | |
| Llama-3.1-Instructtype=off-the-shelf anchor2026.06 | 7.356 | — | — | — | — | — | 7.875 | 6.838 | 1.037 | |
| DPOBackbone=Mistral-7B-Base2026.04 | 7.3 | — | — | — | — | — | — | — | — | |
| SimPOBackbone=Mistral-7B-Base2026.04 | 7.3 | — | — | — | — | — | — | — | — | |
| β-DPOBackbone=Mistral-7B-Base2026.04 | 7.3 | — | — | — | — | — | — | — | — | |
| DivPOModel=Qwen3-4B2026.05 | 7.281 | — | — | — | — | — | — | — | — | |
| 3-level (lin.)training_config=3-level (lin.)2026.06 | 7.269 | — | — | — | — | — | 7.975 | 6.563 | 1.413 | |
| SFTtraining_config=SFT2026.06 | 7.238 | — | — | — | — | — | 7.963 | 6.513 | 1.45 | |
| IPOBackbone=Mistral-7B-Base2026.04 | 7.2 | — | — | — | — | — | — | — | — | |
| AttentionPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 7.19 | — | — | — | — | — | — | — | — | |
| DPOModel=Qwen3-4B2026.05 | 7.113 | — | — | — | — | — | — | — | — | |
| GW-DPOvariant=lin.2026.06 | 7.113 | — | — | — | — | — | 8 | 6.225 | 1.775 | |
| DivPOModel=OLMo-3-7B2026.05 | 7.063 | — | — | — | — | — | — | — | — | |
| REDIPOModel=Qwen3-4B2026.05 | 7.044 | — | — | — | — | — | — | — | — | |
| KTOBackbone=Mistral-7B-Base2026.04 | 7 | — | — | — | — | — | — | — | — | |
| IPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.99 | — | — | — | — | — | — | — | — | |
| REDIPOModel=OLMo-3-7B2026.05 | 6.956 | — | — | — | — | — | — | — | — | |
| InstructModel=OLMo-3-7B2026.05 | 6.919 | — | — | — | — | — | — | — | — | |
| CPOBackbone=Mistral-7B-Base2026.04 | 6.8 | — | — | — | — | — | — | — | — | |
| KTOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.79 | — | — | — | — | — | — | — | — | |
| R-DPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.79 | — | — | — | — | — | — | — | — | |
| BaseReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.78 | — | — | — | — | — | — | — | — | |
| SLiC-HFReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.76 | — | — | — | — | — | — | — | — | |
| ORPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.76 | — | — | — | — | — | — | — | — | |
| DPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.73 | — | — | — | — | — | — | — | — | |
| B4GModel=QWEN2.5-7B, Task=safety enhancement2026.03 | 6.72 | — | — | — | — | — | — | — | — | |
| DPOModel=OLMo-3-7B2026.05 | 6.7 | — | — | — | — | — | — | — | — | |
| CPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.67 | — | — | — | — | — | — | — | — | |
| RRHFReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.63 | — | — | — | — | — | — | — | — | |
| InitialBackbone=Llama3-8B-Base2026.04 | 6.6 | — | — | — | — | — | — | — | — | |
| REDIPOModel=LLaMA-3.1-8B2026.05 | 6.594 | — | — | — | — | — | — | — | — | |
| No-ISEtraining_config=No-ISE2026.06 | 6.544 | — | — | — | — | — | 7.163 | 5.925 | 1.238 | |
| DPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.54 | — | — | — | — | — | — | — | — | |
| AttentionPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.52 | — | — | — | — | — | — | — | — | |
| SimPOReference Model (πref)=LLaMA-3-8B-Instruct2026.05 | 6.51 | — | — | — | — | — | — | — | — | |
| InstructModel=LLaMA-3.1-8B2026.05 | 6.475 | — | — | — | — | — | — | — | — | |
| DPOModel=LLaMA-3.1-8B2026.05 | 6.463 | — | — | — | — | — | — | — | — | |
| B4GModel=QWEN2.5-7B, Task=access control2026.03 | 6.39 | — | — | — | — | — | — | — | — | |
| B4GModel=QWEN2.5-7B, Task=model identity2026.03 | 6.34 | — | — | — | — | — | — | — | — | |
| R-DPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.33 | — | — | — | — | — | — | — | — | |
| B4GModel=LLAMA3.1-8B, Task=safety enhancement2026.03 | 6.31 | — | — | — | — | — | — | — | — | |
| InitialBackbone=Mistral-7B-Base2026.04 | 6.3 | — | — | — | — | — | — | — | — | |
| IPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.23 | — | — | — | — | — | — | — | — | |
| B4GModel=LLAMA3.1-8B, Task=model identity2026.03 | 6.2 | — | — | — | — | — | — | — | — | |
| B4GModel=GEMMA-2-9B, Task=access control2026.03 | 6.2 | — | — | — | — | — | — | — | — | |
| RRHFReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.18 | — | — | — | — | — | — | — | — | |
| B4GModel=LLAMA2-13B, Task=access control2026.03 | 6.14 | — | — | — | — | — | — | — | — | |
| CPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.13 | — | — | — | — | — | — | — | — | |
| SimPOReference Model (πref)=LLaMA-3-8B-Base-SFT2026.05 | 6.13 | — | — | — | — | — | — | — | — |