Dialogue Alignment Evaluation on AlignBench
6.76ReasoningTaP-SFT (GPT-4)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TaP-SFT (GPT-4)Base Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 6.76 | 6.51 | 6.64 | |
| TaP-SFT (GPT-4)Base Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 6.62 | 6.27 | 6.45 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 6.46 | 6.24 | 6.35 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 6.34 | 5.92 | 6.13 | |
| TaP-SFT (GPT-4)Base Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 6.27 | 6.27 | 6.27 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 6.25 | 6.07 | 6.16 | |
| TaP-SFT (GPT-4)Base Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 6.1 | 5.83 | 5.96 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 5.77 | 5.93 | 5.85 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 5.64 | 5.73 | 5.68 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 5.34 | 5.73 | 5.54 | |
| Infinity-InstructBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 5.26 | 5.94 | 5.6 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 5.25 | 5.65 | 5.45 | |
| MOSS-SFTBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 5.25 | 5.59 | 5.42 | |
| Infinity-InstructBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 5.16 | 5.81 | 5.48 | |
| COIG-CQIABase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 5.14 | 5.27 | 5.2 | |
| TaP-SFT (GPT-4)Base Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 4.99 | 5.85 | 5.42 | |
| MOSS-SFTBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 4.96 | 5.48 | 5.22 | |
| COIG-CQIABase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 4.94 | 5.26 | 5.1 | |
| Infinity-InstructBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 4.92 | 5.63 | 5.28 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 4.9 | 5.31 | 5.1 | |
| TaP-SFT (DeepSeek-V2)Base Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 4.89 | 5.39 | 5.14 | |
| COIG-CQIABase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 4.72 | 5.13 | 4.92 | |
| MOSS-SFTBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 4.71 | 5.26 | 4.98 | |
| Infinity-InstructBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 4.7 | 5.54 | 5.12 | |
| BELLE-SFTBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 4.7 | 5.7 | 5.2 | |
| TaP-SFT (GPT-4)Base Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 4.57 | 5.46 | 5.02 | |
| MOSS-SFTBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 4.55 | 5.17 | 4.86 | |
| BELLE-SFTBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 4.55 | 5.61 | 5.08 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 4.53 | 5.14 | 4.83 | |
| COIG-CQIABase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 4.51 | 5.02 | 4.76 | |
| BELLE-SFTBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 4.32 | 5.44 | 4.88 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 4.16 | 4.87 | 4.51 | |
| BELLE-SFTBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 4.16 | 5.35 | 4.76 | |
| TaP-SFT (GPT-4)Base Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 4.1 | 5.17 | 4.64 | |
| FireflyBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 4.1 | 4.89 | 4.49 | |
| Infinity-InstructBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 4.02 | 5.11 | 4.57 | |
| Alpaca-GPT-4-ZHBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.99 | 4.75 | 4.37 | |
| TaP-SFT (DeepSeek-V2)Base Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 3.96 | 4.9 | 4.43 | |
| FireflyBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 3.94 | 4.94 | 4.44 | |
| Infinity-InstructBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.91 | 4.92 | 4.41 | |
| BELLE-SFTBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 3.88 | 5.11 | 4.49 | |
| TaP-SFT (GPT-4)Base Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 3.87 | 5.14 | 4.5 | |
| MOSS-SFTBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 3.82 | 4.96 | 4.39 | |
| BELLE-SFTBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.82 | 4.99 | 4.41 | |
| MOSS-SFTBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.73 | 4.82 | 4.27 | |
| Infinity-InstructBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 3.72 | 5.21 | 4.47 | |
| BELLE-SFTBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 3.72 | 5.16 | 4.44 | |
| BELLE-SFTBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 3.68 | 5.07 | 4.37 | |
| COIG-CQIABase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.65 | 4.31 | 3.98 | |
| FireflyBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 3.62 | 4.64 | 4.13 | |
| COIG-CQIABase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 3.61 | 4.38 | 3.99 | |
| TaP-SFT (DeepSeek-V2)Base Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 3.61 | 4.75 | 4.18 | |
| TaP-SFT (DeepSeek-V2)Base Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 3.61 | 4.58 | 4.1 | |
| TaP-SFT (GPT-4)Base Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 3.6 | 4.88 | 4.24 | |
| Infinity-InstructBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 3.57 | 4.99 | 4.28 | |
| BELLE-SFTBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 3.47 | 5.05 | 4.26 | |
| TaP-SFT (DeepSeek-V2)Base Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 3.47 | 4.61 | 4.04 | |
| FireflyBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 3.45 | 4.61 | 4.03 | |
| TaP-SFT (GPT-4)Base Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 3.4 | 4.73 | 4.06 | |
| FireflyBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 3.38 | 4.37 | 3.87 | |
| COIGBase Model=Qwen2.5-14B, Evaluator=DeepSeek-V32025.06 | 3.38 | 4.86 | 4.12 | |
| COIGBase Model=Qwen2.5-14B, Evaluator=GPT-4o2025.06 | 3.3 | 4.84 | 4.07 | |
| BELLE-SFTBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 3.24 | 4.88 | 4.06 | |
| FireflyBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 3.22 | 4.34 | 3.78 | |
| Alpaca-GPT-4-ZHBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 3.19 | 4.17 | 3.68 | |
| Alpaca-GPT-4-ZHBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 3.17 | 4.22 | 3.7 | |
| Infinity-InstructBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 3.15 | 5.07 | 4.11 | |
| COIGBase Model=Qwen2.5-7B, Evaluator=DeepSeek-V32025.06 | 3.11 | 4.38 | 3.75 | |
| Infinity-InstructBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 3.01 | 4.95 | 3.98 | |
| COIGBase Model=Qwen2.5-3B, Evaluator=DeepSeek-V32025.06 | 2.84 | 4.03 | 3.43 | |
| MOSS-SFTBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 2.84 | 4.41 | 3.63 | |
| MOSS-SFTBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 2.84 | 4.21 | 3.53 | |
| COIGBase Model=Qwen2.5-7B, Evaluator=GPT-4o2025.06 | 2.82 | 4.41 | 3.62 | |
| Alpaca-GPT-4-ZHBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 2.77 | 4.28 | 3.53 | |
| COIGBase Model=Qwen2.5-3B, Evaluator=GPT-4o2025.06 | 2.73 | 4.12 | 3.42 | |
| MOSS-SFTBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 2.73 | 4.19 | 3.46 | |
| MOSS-SFTBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 2.69 | 4.32 | 3.5 | |
| COIG-CQIABase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 2.69 | 3.58 | 3.14 | |
| Alpaca-GPT-4-ZHBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 2.64 | 4.12 | 3.38 | |
| COIG-CQIABase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 2.64 | 3.56 | 3.1 | |
| COIG-CQIABase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 2.54 | 3.57 | 3.06 | |
| FireflyBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 2.44 | 4.06 | 3.25 | |
| FireflyBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 2.4 | 3.97 | 3.19 | |
| FireflyBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 2.4 | 4.07 | 3.24 | |
| COIG-CQIABase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 2.39 | 3.56 | 2.97 | |
| FireflyBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 2.36 | 3.95 | 3.15 | |
| COIGBase Model=Gemma-2-9B, Evaluator=DeepSeek-V32025.06 | 1.95 | 2.8 | 2.38 | |
| COIGBase Model=Llama-3.1-8B, Evaluator=DeepSeek-V32025.06 | 1.88 | 3.04 | 2.46 | |
| COIGBase Model=Llama-3.1-8B, Evaluator=GPT-4o2025.06 | 1.79 | 2.98 | 2.38 | |
| COIGBase Model=Gemma-2-9B, Evaluator=GPT-4o2025.06 | 1.7 | 2.81 | 2.25 |