Mathematical Reasoning on MMLU Math (Accuracy)
61.09AccuracyGREATS
Evaluation results
| Method | Links | |
|---|---|---|
| GREATSBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 61.09 | |
| PartitionSelBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 60.99 | |
| PartitionSelBackbone=Qwen2.5-3B, Fine-tuning protocol=LoRA, Data mixture strategy=PartitionSel2026.06 | 60.57 | |
| IDBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 59.65 | |
| IWDBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 57.95 | |
| GradNormBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 55.65 | |
| COLMBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 55.03 | |
| RandomBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.06 | 53.32 | |
| DoReMiBackbone=Qwen2.5-3B, Fine-tuning protocol=LoRA, Data mixture strategy=DoReMi2026.06 | 53.08 | |
| COLMBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 40.35 | |
| COLMBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 40.35 | |
| GradNormBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 39.53 | |
| GradNormBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 39.53 | |
| PartitionSelBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 39.01 | |
| PartitionSelBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 39.01 | |
| IDBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 36.65 | |
| IDBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.06 | 36.65 |