Cultural Reasoning on CulturalBench-Hard (CB-H) (test)
46.98AccuracyC-Mining
Evaluation Results
| Method | Links | |
|---|---|---|
| C-MiningBackbone=Qwen3-32B, Instruction Tuning Method=C-Mining2026.04 | 46.98 | |
| Qwen3-32BBackbone=Qwen3-32B, Instruction Tuning Method=Standard2026.04 | 44.62 | |
| C-MiningBackbone=Qwen2.5-7B, Instruction Tuning Method=C-Mining2026.04 | 40.78 | |
| CultureBankBackbone=Qwen2.5-7B, Instruction Tuning Method=CultureBank2026.04 | 40.13 | |
| CultureLLMBackbone=Qwen2.5-7B, Instruction Tuning Method=CultureLLM2026.04 | 38.99 | |
| Llama3.1-8BBackbone=Llama3.1-8B, Instruction Tuning Method=Standard2026.04 | 37.44 | |
| Qwen2.5-7BBackbone=Qwen2.5-7B, Instruction Tuning Method=Standard2026.04 | 34.75 | |
| GLM4-9BBackbone=GLM4-9B, Instruction Tuning Method=Standard2026.04 | 34.1 | |
| CultureSPABackbone=Qwen2.5-7B, Instruction Tuning Method=CultureSPA2026.04 | 30.18 | |
| Ministral3-8BBackbone=Ministral3-8B, Instruction Tuning Method=Standard2026.04 | 26.67 |