Cultural Alignment on WorldValuesBench (WVB)
50.64AccuracyCUMA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CUMABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=4.15%, Rank (r)=642026.01 | 50.64 | 31.5 | 0.1876 | |
| CUMABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=4.15%, Rank (r)=642026.01 | 50.46 | 32.5 | 0.187 | |
| CUMABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=1.53%, Rank (r)=82026.01 | 49.02 | 29.7 | 0.198 | |
| CUMABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=1.53%, Rank (r)=82026.01 | 48.9 | 30.5 | 0.1903 | |
| HydraLoRABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=2.31%2026.01 | 46.5 | 29.9 | 0.235 | |
| Full Fine-Tuning (FFT)Backbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=100.0%2026.01 | 45.54 | 28.21 | 0.2228 | |
| HydraLoRABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=2.31%2026.01 | 45.36 | 28.12 | 0.2793 | |
| Full Fine-Tuning (FFT)Backbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=100.0%2026.01 | 45.25 | 30.5 | 0.2205 | |
| MixLoRABackbone=Llama-3.1-8B, Category=Sparsely Activated Adapters, Trainable Params=3.01%2026.01 | 45.2 | 29.8 | 0.244 | |
| P-Tuning v2Backbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.94%2026.01 | 45.04 | 28.17 | 0.2358 | |
| P-Tuning v2Backbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.94%2026.01 | 43.8 | 29.1 | 0.247 | |
| MixLoRABackbone=Qwen3-8B, Category=Sparsely Activated Adapters, Trainable Params=3.01%2026.01 | 43.5 | 26.44 | 0.2904 | |
| DoRABackbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.38%, Rank (r)=642026.01 | 42.78 | 24.73 | 0.2773 | |
| LoRABackbone=Qwen3-8B, Category=Dense Fine-Tuning, Trainable Params=0.37%, Rank (r)=642026.01 | 40.06 | 22.02 | 0.27 | |
| Persona PromptingBackbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 37.06 | 23.9 | 0.3105 | |
| DoRABackbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.38%, Rank (r)=642026.01 | 36.5 | 25.1 | 0.2587 | |
| Persona PromptingBackbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 34.92 | 21.05 | 0.2864 | |
| LoRABackbone=Llama-3.1-8B, Category=Dense Fine-Tuning, Trainable Params=0.37%, Rank (r)=642026.01 | 34.3 | 22.37 | 0.2537 | |
| Vanilla BaselineBackbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 32.42 | 22.99 | 0.3967 | |
| Vanilla BaselineBackbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 31.68 | 18.92 | 0.3851 | |
| Prompt Steering (3-shot)Backbone=Qwen3-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 28.08 | 12.36 | 0.2299 | |
| Prompt Steering (3-shot)Backbone=Llama-3.1-8B, Category=Inference-Time Strategies, Trainable Params=0.00%2026.01 | 27.5 | 11.14 | 0.2507 |