Emotional Support Conversation on EmoHarbor
3.19Problem ResolutionGPT-5-2025-08-07
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5-2025-08-07Model Type=Close-Source LLM2026.03 | 3.19 | 3.06 | 3.83 | |
| GPT-4o-2024-11-20Model Type=Close-Source LLM2026.03 | 2.56 | 2.17 | 3.2 | |
| Llama-3-8b + RAPOTraining Strategy=RAPO2026.03 | 2.46 | 2.05 | 3.04 | |
| Llama-3-8b + CPOTraining Strategy=CPO2026.03 | 2.4 | 2.03 | 3.03 | |
| PsyChat-Qwen2.5-7BModel Type=Open-Source In-domain LLM2026.03 | 2.03 | 1.54 | 2.55 | |
| Qwen-3-8b + RAPOTraining Strategy=RAPO2026.03 | 2.03 | 1.73 | 2.67 | |
| Llama-3-8bTraining Strategy=Base2026.03 | 2.03 | 2 | 2.99 | |
| Llama-3-8b + GRPOTraining Strategy=GRPO2026.03 | 2.03 | 1.56 | 2.59 | |
| Qwen-3-8b + GRPOTraining Strategy=GRPO2026.03 | 1.99 | 1.57 | 2.58 | |
| Qwen-3-8b + CPOTraining Strategy=CPO2026.03 | 1.99 | 1.67 | 2.58 | |
| Qwen-3-8bTraining Strategy=Base2026.03 | 1.89 | 1.62 | 2.59 | |
| SoulChat2.0-Llama-3.1-8BModel Type=Open-Source In-domain LLM2026.03 | 1.88 | 1.27 | 2.37 | |
| Qwen-2.5-7b + RAPOTraining Strategy=RAPO2026.03 | 1.86 | 1.26 | 2.27 | |
| SoulChat2.0-Qwen2-7BModel Type=Open-Source In-domain LLM2026.03 | 1.84 | 1.2 | 2.29 | |
| Qwen-2.5-7b + CPOTraining Strategy=CPO2026.03 | 1.76 | 1.25 | 2.15 | |
| Qwen-2.5-7bTraining Strategy=Base2026.03 | 1.69 | 1.21 | 2.1 | |
| Doubao-Pro-32kModel Type=Close-Source LLM2026.03 | 1.65 | 1.25 | 2.12 | |
| Qwen-2.5-7b + GRPOTraining Strategy=GRPO2026.03 | 1.6 | 1.04 | 2.02 | |
| Qwen-3-8b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 1.29 | 1.1 | 1.72 | |
| Llama-3-8b + SFT-FullTraining Strategy=SFT-Full2026.03 | 1.25 | 1 | 1.66 | |
| Qwen-2.5-7b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 1.22 | 1.06 | 1.72 | |
| Llama-3-8b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 1.2 | 1.07 | 1.69 | |
| Qwen-2.5-7b + SFT-FullTraining Strategy=SFT-Full2026.03 | 1.13 | 1 | 1.58 | |
| Qwen-3-8b + SFT-FullTraining Strategy=SFT-Full2026.03 | 1.09 | 1 | 1.54 |