Regression on Macro-average SICKR-STS, STS-B, WMT_RU_EN, WMT_EN_ZH, WMT_SI_EN (test)
76.3Pearson Correlation (r)RELISH
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RELISHBackbone Protocol=Fine-tuned, Method Variant=RELISH2026.04 | 76.3 | 74 | 13.3 | |
| RAFTBackbone Protocol=Fine-tuned, Method Variant=RAFT2026.04 | 71.5 | 69 | 16.6 | |
| L+RAILBackbone Protocol=Fine-tuned, Fine-tuning Strategy=LoRA, Method Variant=RAIL2026.04 | 66.9 | 66.1 | 19 | |
| L+MLPBackbone Protocol=Fine-tuned, Fine-tuning Strategy=LoRA, Method Variant=MLP2026.04 | 64.8 | 61.7 | 16.7 | |
| MLPBackbone Protocol=Frozen, Method Variant=MLP2026.04 | 64.7 | 61.7 | 16.7 | |
| L+ZeroBackbone Protocol=Fine-tuned, Fine-tuning Strategy=LoRA, Method Variant=Zero-shot2026.04 | 61.7 | 58.9 | 19.8 | |
| RAILBackbone Protocol=Frozen, Method Variant=RAIL2026.04 | 61.6 | 64.1 | 27.3 | |
| ZeroBackbone Protocol=Frozen, Method Variant=Zero-shot2026.04 | 61 | 60.9 | 26.9 | |
| ManyBackbone Protocol=Frozen, Method Variant=Many-shot2026.04 | 57.8 | 56.1 | 23.9 | |
| L+Lin.Backbone Protocol=Fine-tuned, Fine-tuning Strategy=LoRA, Method Variant=Linear2026.04 | 53.4 | 51 | 24.3 | |
| Lin.Backbone Protocol=Frozen, Method Variant=Linear2026.04 | 52.9 | 50.3 | 24.4 |