Multi-session Dialogue Generation on REALTALK two-session (test)
9.87PPLC-DIC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| C-DICBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 9.87 | 0.034 | 0.132 | 0.175 | 0.029 | |
| AutoCompressorBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 11.15 | 0.02 | 0.112 | 0.146 | 0.022 | |
| TruncationBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 21.865 | 0.026 | 0.109 | 0.174 | 0.028 | |
| SummarizationBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 26.12 | 0.023 | 0.118 | 0.172 | 0.025 | |
| Full promptingBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 27.225 | 0.022 | 0.109 | 0.159 | 0.02 | |
| In-Session RAGBackbone=Llama-2-Chat-7B, Context=Two sessions2026.06 | 27.435 | 0.019 | 0.1 | 0.145 | 0.014 | |
| ICAE (incremental)Backbone=Llama-2-Chat-7B, Context=Two sessions, Mode=incremental2026.06 | 218.103 | 0.017 | 0.059 | 0.073 | 0.007 |