Emotional Support Conversation on ESConv (BERT-SCORE)
85.89BERT-SCOREQwen-3-8b + SFT-Full
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen-3-8b + SFT-FullTraining Strategy=SFT-Full2026.03 | 85.89 | |
| Qwen-2.5-7b + SFT-FullTraining Strategy=SFT-Full2026.03 | 85.85 | |
| Llama-3-8b + SFT-FullTraining Strategy=SFT-Full2026.03 | 85.84 | |
| Llama-3-8b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 85.84 | |
| Qwen-3-8b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 85.8 | |
| SoulChat2.0-Qwen2-7BModel Type=Open-Source In-domain LLM2026.03 | 85.75 | |
| Qwen-2.5-7bTraining Strategy=Base2026.03 | 85.67 | |
| SoulChat2.0-Llama-3.1-8BModel Type=Open-Source In-domain LLM2026.03 | 85.65 | |
| Qwen-2.5-7b + RAPOTraining Strategy=RAPO2026.03 | 85.65 | |
| Qwen-2.5-7b + SFT-HDSTraining Strategy=SFT-HDS2026.03 | 85.61 | |
| Qwen-2.5-7b + CPOTraining Strategy=CPO2026.03 | 85.6 | |
| Qwen-2.5-7b + GRPOTraining Strategy=GRPO2026.03 | 85.57 | |
| Qwen-3-8b + RAPOTraining Strategy=RAPO2026.03 | 85.56 | |
| Llama-3-8b + RAPOTraining Strategy=RAPO2026.03 | 85.56 | |
| Llama-3-8bTraining Strategy=Base2026.03 | 85.47 | |
| Llama-3-8b + GRPOTraining Strategy=GRPO2026.03 | 85.46 | |
| Llama-3-8b + CPOTraining Strategy=CPO2026.03 | 85.33 | |
| Doubao-Pro-32kModel Type=Close-Source LLM2026.03 | 85.11 | |
| Qwen-3-8b + CPOTraining Strategy=CPO2026.03 | 85.01 | |
| PsyChat-Qwen2.5-7BModel Type=Open-Source In-domain LLM2026.03 | 84.99 | |
| GPT-4o-2024-11-20Model Type=Close-Source LLM2026.03 | 84.89 | |
| Qwen-3-8b + GRPOTraining Strategy=GRPO2026.03 | 84.48 | |
| Qwen-3-8bTraining Strategy=Base2026.03 | 84.42 | |
| GPT-5-2025-08-07Model Type=Close-Source LLM2026.03 | 84.26 |