Utterance Generation on ESC (test)
18.75BLEU-1All-In-One + Reasoning + RL
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| All-In-One + Reasoning + RLBase Model=LLaMA3.1-8b, reasoning=true, RL=true2026.04 | 18.75 | 8.62 | 4.07 | 25.68 | 6.26 | 20.71 | 20.68 | |
| All-In-One + ReasoningBase Model=LLaMA3.1-8b, reasoning=true2026.04 | 18.05 | 8.39 | 3.98 | 24.96 | 6.19 | 20.1 | 19.98 | |
| One-by-One + ReasoningBase Model=LLaMA3.1-8b, reasoning=true2026.04 | 17.76 | 8.28 | 3.88 | 24.91 | 6.06 | 20.23 | 20.72 | |
| One-by-One + Reasoning + RLBase Model=LLaMA3.1-8b, reasoning=true, RL=true2026.04 | 17.76 | 8.14 | 3.71 | 25.01 | 5.98 | 20.38 | 21.11 | |
| All-In-OneBase Model=LLaMA3.1-8b2026.04 | 17.04 | 7.19 | 3.06 | 23.33 | 4.74 | 18.27 | 18.17 | |
| One-by-OneBase Model=LLaMA3.1-8b2026.04 | 16.97 | 7.49 | 3.3 | 24.02 | 5.23 | 19.26 | 18.95 | |
| Single StrategyBase Model=LLaMA3.1-8b2026.04 | 16.54 | 6.98 | 3.01 | 22.95 | 4.63 | 18.06 | 18.16 | |
| DeepSeek-R1Model Group=Instruction-following LLMs2026.04 | 12.67 | 4.14 | 1.57 | 16.25 | 2.1 | 14.06 | 15.03 | |
| GPT-5Model Group=Instruction-following LLMs2026.04 | 12.5 | 3.79 | 1.06 | 17.88 | 2.01 | 15.29 | 11.25 | |
| Qwen3-235b-InstructModel Group=Instruction-following LLMs2026.04 | 11.72 | 3.53 | 1.37 | 15.7 | 1.58 | 13.51 | 12.44 |