Multitask Language Understanding on MMLU & SuperGLUE (test)
51.6MMLU AccuracyPEML
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PEMLBackbone=LLaMA-7B, n=1, r=96, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 51.6 | 92.3 | 92.8 | 89.6 | 75.5 | 80.3 | |
| MultiLoRABackbone=LLaMA-7B, n=5, r=32, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 51.4 | 88.5 | 89.4 | 89.4 | 71.4 | 78 | |
| MultiLoRABackbone=LLaMA-7B, n=3, r=32, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 51.2 | 87.8 | 88.7 | 89.7 | 70.8 | 77.6 | |
| LoRABackbone=LLaMA-7B, n=1, r=160, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 50.2 | 87.7 | 85.3 | 83.3 | 70.1 | 75.3 | |
| FTBackbone=LLaMA-7B, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 49.5 | 88.4 | 87.2 | 85.2 | 74 | 76.9 | |
| LoRABackbone=LLaMA-7B, n=1, r=96, MMLU Evaluation=5-shot, SuperGLUE Evaluation=zero-shot2026.05 | 47.7 | 88.2 | 85.4 | 83.4 | 71.6 | 75.2 |