Cross-lingual Cultural Consistency on BLEnD Higher-Resource
0.025Max SigmaVanilla
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VanillaModel=Gemma-2-27B2026.04 | 0.025 | 0.625 | — | — | — | |
| Few-shotModel=Gemma-2-27B2026.04 | 0.025 | 0.631 | — | — | — | |
| SteeringModel=Gemma-2-27B2026.04 | 0.025 | — | 0.599 | 0.608 | 0.614 | |
| PersonaModel=Gemma-2-27B2026.04 | 0.025 | — | 0.619 | 0.655 | 0.684 | |
| C-3POModel=Gemma-2-27B2026.04 | 0.025 | 0.661 | — | — | — | |
| VanillaModel=Llama-3.1-8B2026.04 | 0.026 | 0.469 | — | — | — | |
| Few-shotModel=Llama-3.1-8B2026.04 | 0.026 | 0.501 | — | — | — | |
| SteeringModel=Llama-3.1-8B2026.04 | 0.026 | — | 0.459 | 0.468 | 0.475 | |
| PersonaModel=Llama-3.1-8B2026.04 | 0.026 | — | 0.419 | 0.464 | 0.551 | |
| C-3POModel=Llama-3.1-8B2026.04 | 0.026 | 0.532 | — | — | — | |
| VanillaModel=Qwen-2.5-3B2026.04 | 0.027 | 0.461 | — | — | — | |
| Few-shotModel=Qwen-2.5-3B2026.04 | 0.027 | 0.488 | — | — | — | |
| SteeringModel=Qwen-2.5-3B2026.04 | 0.027 | — | 0.454 | 0.463 | 0.471 | |
| PersonaModel=Qwen-2.5-3B2026.04 | 0.027 | — | 0.402 | 0.446 | 0.463 | |
| C-3POModel=Qwen-2.5-3B2026.04 | 0.027 | 0.58 | — | — | — |