Open-ended Evaluation on DomainBench (test)
27.98Geography ScoreTF-TTCL
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TF-TTCLBackbone=Llama-3.1-8B-Instruct, Evaluation Protocol=test-time adaptation2026.04 | 27.98 | 10.95 | 20.18 | 28.63 | 21.94 | |
| EATABackbone=Llama-3.1-8B-Instruct, Publication=ICML 2022, Evaluation Protocol=test-time adaptation2026.04 | 27.57 | 6.26 | 14.55 | 18.86 | 16.81 | |
| TentBackbone=Llama-3.1-8B-Instruct, Publication=ICLR 2021, Evaluation Protocol=test-time adaptation2026.04 | 26.82 | 6.24 | 14.48 | 21.4 | 17.24 | |
| COMEBackbone=Llama-3.1-8B-Instruct, Publication=ICLR 2025, Evaluation Protocol=test-time adaptation2026.04 | 26.36 | 4.07 | 13.82 | 6.99 | 12.81 | |
| TLMBackbone=Llama-3.1-8B-Instruct, Publication=ICML 2025, Evaluation Protocol=test-time adaptation2026.04 | 26.2 | 9.56 | 13.72 | 22.95 | 18.11 | |
| Base LLMBackbone=Llama-3.1-8B-Instruct, Evaluation Protocol=zero-shot2026.04 | 24.41 | 8.76 | 13.56 | 22.51 | 17.31 | |
| TF-GRPOBackbone=Llama-3.1-8B-Instruct, Publication=arXiv 2025, Evaluation Protocol=test-time adaptation2026.04 | 22.6 | 9.93 | 11.47 | 20.71 | 16.18 |