Text Summarization on CNN/DM (Hal. Rate, ROUGE-L)
20.5Hal. RateSafeLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SafeLMsetup=full2026.04 | 20.5 | 44.3 | |
| RAG only2026.04 | 21.3 | 44.1 | |
| SafeLMsetup=no MG2026.04 | 23.1 | 43.9 | |
| Vanilla fine-tune2026.04 | 34.7 | 42.8 |
Loading the SOTA2 catalog…