Multi-Turn Dialogue Generation on REALTALK zero-shot
9.789PPLC-DIC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| C-DICBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 9.789 | 0.035 | 0.134 | 0.176 | 0.03 | |
| AutoCompressorBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 12.625 | 0.019 | 0.055 | 0.134 | 0.019 | |
| ICAEBackbone=Llama-2-Chat-7B, Compression strategy=one-shot, Zero-shot evaluation=true2026.06 | 21.39 | 0.025 | 0.118 | 0.166 | 0.026 | |
| TruncationBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 23.83 | 0.023 | 0.114 | 0.165 | 0.022 | |
| Full promptingBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 25.546 | 0.022 | 0.11 | 0.16 | 0.02 | |
| SummarizationBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 26.087 | 0.023 | 0.114 | 0.168 | 0.023 | |
| RAGBackbone=Llama-2-Chat-7B, RAG top-k=5, Zero-shot evaluation=true2026.06 | 26.789 | 0.02 | 0.103 | 0.151 | 0.015 | |
| RAGBackbone=Llama-2-Chat-7B, RAG top-k=10, Zero-shot evaluation=true2026.06 | 27.311 | 0.019 | 0.102 | 0.148 | 0.014 | |
| RAGBackbone=Llama-2-Chat-7B, RAG top-k=20, Zero-shot evaluation=true2026.06 | 27.646 | 0.019 | 0.102 | 0.148 | 0.014 | |
| RAG-thresholdBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 28.238 | 0.009 | 0.101 | 0.147 | 0.013 | |
| Full promptingBackbone=Llama3.1-8B, Zero-shot evaluation=true2026.06 | 29.444 | 0.017 | 0.109 | 0.158 | 0.018 | |
| InfLLMBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 29.995 | 0.013 | 0.099 | 0.141 | 0.014 | |
| LLMLinguaBackbone=Llama-2-Chat-7B, Zero-shot evaluation=true2026.06 | 31.067 | 0.008 | 0.092 | 0.131 | 0.011 | |
| ICAEBackbone=Llama-2-Chat-7B, Compression strategy=incremental, Zero-shot evaluation=true2026.06 | 124.024 | 0.02 | 0.068 | 0.086 | 0.013 |