Knowledge Synthesis on MMLU
59.25AUROCMATU
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MATULLM=GPT-4o2026.04 | 59.25 | 51.52 | |
| MATULLM=Qwen2.5-7B2026.04 | 58.21 | 77.68 | |
| Eigv(Agre)-WholeLLM=GPT-4o2026.04 | 56.41 | 50.49 | |
| P(true)LLM=GPT-4o2026.04 | 55.94 | 49.76 | |
| Eigv(Agre)-finalLLM=GPT-4o2026.04 | 55.68 | 49.52 | |
| P(true)LLM=Qwen2.5-7B2026.04 | 55.52 | 76.81 | |
| MATULLM=Llama3.1-8B2026.04 | 55 | 67.97 | |
| Eigv(Agre)-finalLLM=Qwen2.5-7B2026.04 | 54.46 | 76.5 | |
| SAUP-MultipleLLM=Qwen2.5-7B2026.04 | 53.82 | 76.02 | |
| SAUP-MultipleLLM=Llama3.1-8B2026.04 | 53.82 | 67.19 | |
| SAUP-SingleLLM=Llama3.1-8B2026.04 | 53.4 | 65.42 | |
| Eigv(Agre)-WholeLLM=Qwen2.5-7B2026.04 | 53.37 | 74.33 | |
| Eigv(Agre)-finalLLM=Llama3.1-8B2026.04 | 53.21 | 65.86 | |
| P(true)LLM=Llama3.1-8B2026.04 | 52.97 | 66.32 | |
| Eigv(Agre)-WholeLLM=Llama3.1-8B2026.04 | 52.15 | 65.12 | |
| SAUP-SingleLLM=Qwen2.5-7B2026.04 | 50.48 | 72.61 |