Multi-turn Conversation on MT-Bench (AVG Metric)
85.25Average ScoreCoDIT-Qwen3-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoDIT-Qwen3-8BStudent Model=Qwen3-8B-Base2026.04 | 85.25 | — | — | |
| DEPOIteration=32026.05 | 84.5 | — | — | |
| HRC+SPPOITER=2, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 84.2 | 86.1 | 82.4 | |
| CoDIT-Qwen3-30BStudent Model=Qwen3-8B-Base2026.04 | 84.18 | — | — | |
| POPOIteration=32026.05 | 83.8 | — | — | |
| GPM+SPPOITER=2, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 83.6 | 86.1 | 81 | |
| CoDIT-Gemma3Student Model=Qwen3-8B-Base2026.04 | 83.13 | — | — | |
| DPOIteration=32026.05 | 82.9 | — | — | |
| HRC+DSPPOITER=2, Preference Model=GEMMA-2B-IT2026.05 | 82.9 | 84.9 | 81 | |
| WebR-ProStudent Model=Qwen3-8B-Base2026.04 | 82.72 | — | — | |
| BT+SPPOITER=1, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 82.5 | 84.2 | 80.8 | |
| HRC+DSPPOITER=3, Preference Model=GEMMA-2B-IT2026.05 | 82.4 | 86.1 | 78.6 | |
| XPOIteration=32026.05 | 82.1 | — | — | |
| GPM+SPPOITER=1, Preference Model=GEMMA-2B-IT2026.05 | 82.1 | 85.8 | 78.5 | |
| HRC+SPPOITER=1, Preference Model=GEMMA-2B-IT2026.05 | 82.1 | 84.5 | 79.8 | |
| HRC+SPPOITER=3, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 82.1 | 84.8 | 79.4 | |
| HRC+SPPOITER=2, Preference Model=GEMMA-2B-IT2026.05 | 81.8 | 83.9 | 79.6 | |
| BT+SPPOITER=2, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 81.6 | 83.9 | 79.2 | |
| HRC+DSPPOITER=2, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 81.4 | 85.5 | 77.4 | |
| HRC+DSPPOITER=1, Preference Model=GEMMA-2B-IT2026.05 | 81.1 | 84.8 | 77.5 | |
| HRC+DSPPOITER=1, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 81.1 | 84.6 | 77.5 | |
| HRC+SPPOITER=3, Preference Model=GEMMA-2B-IT2026.05 | 81.1 | 85 | 77.2 | |
| BT+SPPOITER=3, Preference Model=GEMMA-2B-IT2026.05 | 80.8 | 84.8 | 76.9 | |
| BASEITER=-, Preference Model=GEMMA-2B-IT2026.05 | 80.7 | 83.9 | 77.5 | |
| BASEITER=-, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 80.7 | 83.9 | 77.5 | |
| Llama-3.1-LMSYS-Chat-1M-SynthStudent Model=Qwen3-8B-Base2026.04 | 80.05 | — | — | |
| BT+SPPOITER=1, Preference Model=GEMMA-2B-IT2026.05 | 80 | 83.8 | 76.2 | |
| BT+SPPOITER=3, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 79.9 | 82.1 | 77.8 | |
| BT+SPPOITER=2, Preference Model=GEMMA-2B-IT2026.05 | 79.8 | 82.1 | 77.4 | |
| GPM+SPPOITER=2, Preference Model=GEMMA-2B-IT2026.05 | 79.8 | 83 | 76.5 | |
| GPM+SPPOITER=1, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 79 | 83.8 | 74.2 | |
| HRC+DSPPOITER=3, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 78.6 | 80.4 | 76.8 | |
| HRC+SPPOITER=1, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 77.8 | 83.9 | 71.6 | |
| GPM+SPPOITER=3, Preference Model=GEMMA-2B-IT2026.05 | 77 | 79.8 | 74.2 | |
| Llama-3-8B-SFT2026.05 | 76.6 | — | — | |
| DomLoRABase Model=Qwen3-8B, Params=2.1M2026.05 | 75.9 | — | — | |
| GPM+SPPOITER=3, Preference Model=LLAMA-3.1-8B-INSTRUCT2026.05 | 75.5 | 78.5 | 72.5 | |
| WebR-BasicStudent Model=Qwen3-8B-Base2026.04 | 75.43 | — | — | |
| CoDIT-Qwen3-8BStudent Model=Llama-3.1-8B2026.04 | 75.39 | — | — | |
| Gemma-2-LMSYS-Chat-1M-SynthStudent Model=Qwen3-8B-Base2026.04 | 74.56 | — | — | |
| CoDIT-Gemma3Student Model=Llama-3.1-8B2026.04 | 74.33 | — | — | |
| Magpie-Pro-300K-FilteredStudent Model=Qwen3-8B-Base2026.04 | 74.05 | — | — | |
| CoDIT-Qwen3-30BStudent Model=Llama-3.1-8B2026.04 | 73.73 | — | — | |
| Llama-3.1-LMSYS-Chat-1M-SynthStudent Model=Llama-3.1-8B2026.04 | 72.11 | — | — | |
| WildChatStudent Model=Qwen3-8B-Base2026.04 | 70.04 | — | — | |
| WebR-ProStudent Model=Llama-3.1-8B2026.04 | 70 | — | — | |
| Gemma-2-LMSYS-Chat-1M-SynthStudent Model=Llama-3.1-8B2026.04 | 69.51 | — | — | |
| PLoPBase Model=Qwen3-8B, Params=131M2026.05 | 69.1 | — | — | |
| DomLoRABase Model=LLaMA-3.1-8B, Params=2.3M2026.05 | 67 | — | — | |
| WebR-BasicStudent Model=Llama-3.1-8B2026.04 | 64.31 | — | — | |
| ISTBase Model=Qwen3-8B, Params=334M2026.05 | 63.2 | — | — | |
| Magpie-Pro-300K-FilteredStudent Model=Llama-3.1-8B2026.04 | 63.15 | — | — | |
| LoRABase Model=Qwen3-8B, Params=334M2026.05 | 62.5 | — | — | |
| WildChatStudent Model=Llama-3.1-8B2026.04 | 62.24 | — | — | |
| ISTBase Model=LLaMA-3.1-8B, Params=321M2026.05 | 62.1 | — | — | |
| LoRABase Model=LLaMA-3.1-8B, Params=321M2026.05 | 61.6 | — | — | |
| PLoPBase Model=LLaMA-3.1-8B, Params=125M2026.05 | 60.8 | — | — | |
| VanillaBackbone=Qwen2.5-7B-Instruct2026.05 | 8.82 | — | — | |
| I-DPO +MaPPOBackbone=Qwen2.5-14B-Instruct2025.07 | 8.66 | — | — | |
| GRPOBackbone=Qwen2.5-14B-Instruct, clip ratio=0.22025.07 | 8.65 | — | — | |
| PPOBackbone=Qwen2.5-14B-Instruct, clip ratio=0.22025.07 | 8.61 | — | — | |
| EVA (text)Backbone=Qwen2.5-7B-Instruct2026.05 | 8.48 | — | — | |
| LoRABackbone=Qwen2.5-7B-Instruct2026.05 | 8.39 | — | — | |
| SafeDecodingBackbone=Qwen2.5-7B-Instruct2026.05 | 8.32 | — | — | |
| S-SPPOBackbone=Llama-3-8B, Iteration=Iter32026.06 | 8.22 | 8.54 | 7.9 | |
| S-SPPOBackbone=Llama-3-8B, Iteration=Iter22026.06 | 8.21 | 8.6 | 7.82 | |
| LoRABackbone=Llama3.1-8B-Instruct2026.05 | 8.14 | — | — | |
| SafeDecodingBackbone=Llama3.1-8B-Instruct2026.05 | 8.05 | — | — | |
| S-SPPOBackbone=Llama-3-8B, Iteration=Iter12026.06 | 8.03 | 8.2 | 7.85 | |
| Llama-3-8B-SPPOBackbone=Llama-3-8B, Iteration=Iter12026.06 | 8.01 | 8.29 | 7.73 | |
| Llama-3-8B-SPPOBackbone=Llama-3-8B, Iteration=Iter22026.06 | 7.97 | 8.33 | 7.61 | |
| VanillaBackbone=Mistral-7B-Instruct-v0.22026.05 | 7.93 | — | — | |
| Llama-3-8B-SPPOBackbone=Llama-3-8B, Iteration=Iter32026.06 | 7.93 | 8.36 | 7.49 | |
| Circuit BreakersBackbone=Qwen2.5-7B-Instruct2026.05 | 7.91 | — | — | |
| LEDBackbone=Qwen2.5-7B-Instruct2026.05 | 7.87 | — | — | |
| VanillaBackbone=Llama3.1-8B-Instruct2026.05 | 7.79 | — | — | |
| S-SPPOBackbone=Mistral-7B, Iteration=Iter32026.06 | 7.66 | 7.96 | 7.36 | |
| Mistral-7B-SPPOBackbone=Mistral-7B, Iteration=Iter32026.06 | 7.59 | 7.84 | 7.34 | |
| Snorkel (Mistral-PairRM-DPO)Backbone=Mistral-7B2026.06 | 7.58 | 7.83 | 7.33 | |
| S-SPPOBackbone=Mistral-7B, Iteration=Iter12026.06 | 7.55 | 7.84 | 7.26 | |
| S-SPPOBackbone=Mistral-7B, Iteration=Iter22026.06 | 7.55 | 8.04 | 7.05 | |
| LoRABackbone=Mistral-7B-Instruct-v0.22026.05 | 7.54 | — | — | |
| Mistral-7B-Instruct-v0.2Backbone=Mistral-7B2026.06 | 7.51 | 7.78 | 7.25 | |
| Mistral-7B-SPPOBackbone=Mistral-7B, Iteration=Iter22026.06 | 7.49 | 7.9 | 7.08 | |
| EVA (text)Backbone=Llama3.1-8B-Instruct2026.05 | 7.44 | — | — | |
| EVA (text)Backbone=Mistral-7B-Instruct-v0.22026.05 | 7.35 | — | — | |
| LEDBackbone=Llama3.1-8B-Instruct2026.05 | 7.29 | — | — | |
| Circuit BreakersBackbone=Llama3.1-8B-Instruct2026.05 | 7.21 | — | — | |
| Mistral-7B-SPPOBackbone=Mistral-7B, Iteration=Iter12026.06 | 7.21 | 7.63 | 6.79 | |
| SafeDecodingBackbone=Mistral-7B-Instruct-v0.22026.05 | 7.16 | — | — | |
| LEDBackbone=Mistral-7B-Instruct-v0.22026.05 | 7.09 | — | — | |
| Mistral-7B-DPOBackbone=Mistral-7B, Iteration=Iter32026.06 | 7.09 | 7.49 | 6.69 | |
| Mistral-7B-DPOBackbone=Mistral-7B, Iteration=Iter22026.06 | 7.06 | 7.57 | 6.56 | |
| Mistral-7B-DPOBackbone=Mistral-7B, Iteration=Iter12026.06 | 7.02 | 7.45 | 6.58 | |
| LoRABackbone=Llama2-7B-chat2026.05 | 6.9 | — | — | |
| VanillaBackbone=Llama2-7B-chat2026.05 | 6.89 | — | — | |
| EVA (text)Backbone=Vicuna-7B-v1.52026.05 | 6.84 | — | — | |
| VanillaBackbone=Vicuna-7B-v1.52026.05 | 6.77 | — | — | |
| SafeDecodingBackbone=Vicuna-7B-v1.52026.05 | 6.61 | — | — | |
| EVA (text)Backbone=Llama2-7B-chat2026.05 | 6.31 | — | — |