General Instruction Tuning on MMLU, TyDiQA, CQA, TruthfulQA, GSM8K, and LogiQA
71.4MMLUDomLoRA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DomLoRAModel=Qwen3-8B, Params=2.1M2026.05 | 71.4 | 71.4 | 82.6 | 72.6 | 92.6 | 56.5 | 74.5 | |
| LoRAModel=Qwen3-8B, Params=334M2026.05 | 71 | 68.8 | 81.3 | 73.3 | 87 | 56.2 | 72.9 | |
| PLoPModel=Qwen3-8B, Params=131M2026.05 | 70.9 | 70.8 | 82.1 | 71.9 | 89.8 | 57.3 | 73.8 | |
| ISTModel=Qwen3-8B, Params=334M2026.05 | 70.3 | 69 | 80.8 | 74.5 | 91.3 | 56.8 | 73.8 | |
| DomLoRAModel=LLaMA-3.1-8B, Params=2.3M2026.05 | 64.1 | 69.9 | 73.5 | 54.4 | 85.5 | 41.2 | 64.8 | |
| LoRAModel=LLaMA-3.1-8B, Params=321M2026.05 | 63.5 | 67.2 | 69.8 | 45.2 | 81.4 | 37.6 | 60.8 | |
| ISTModel=LLaMA-3.1-8B, Params=321M2026.05 | 62.5 | 69.1 | 73.7 | 38.8 | 82.5 | 36.9 | 60.6 | |
| PLoPModel=LLaMA-3.1-8B, Params=125M2026.05 | 61.8 | 69.6 | 70.9 | 44.7 | 81.3 | 35.3 | 60.6 |