Summarization on Legal (OOV_SD)
25.16R-LCSCPTOnly (No Vocab Adapt)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CPTOnly (No Vocab Adapt)Backbone=Qwen2.5-7B, Best Ckpts.=10500, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 25.16 | 1.11 | 1.1 | 67.6 | |
| VOCABADAPTBackbone=Llama-3.1-8B, Best Ckpts.=6500, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 24.89 | 1.01 | 1.01 | 68.12 | |
| VOCABADAPTBestsetting=zero-shot2026.05 | 24.89 | — | — | 68.12 | |
| CPTOnly (No Vocab Adapt)Backbone=Llama-3.1-8B, Best Ckpts.=10000, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 24.83 | 1.08 | 1.06 | 68.04 | |
| Llama-3.1-8B-BASEBest Ckpts.=-, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 24.82 | 1.08 | 1.06 | 64.1 | |
| VOCABADAPTBackbone=Qwen2.5-7B, Best Ckpts.=6500, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 23.6 | 1.05 | 1.05 | 66.19 | |
| GPT-5-minisetting=zero-shot2026.05 | 19.45 | — | — | 66.4 | |
| Qwen2.5-7B-BASEBest Ckpts.=-, Evaluation Protocol=In-context learning, Demonstration=One exemplar2026.05 | 9.97 | 1.11 | 1.1 | 27.64 |