List your company
ResearchFormerly WizWand

Multi-task Language Understanding on MMLU (Category Accuracy Breakdown)

68.71STEM AccuracyLoRA + MoLS
24.92636.29347.6659.027May 7, 2026
Updated 5mo ago

Evaluation results

MethodLinks
LoRA + MoLSBackbone=Qwen2.5-7B2026.0568.7182.8766.4177.2173.01
APOLLOBackbone=Qwen2.5-7B2026.0568.4982.9765.876.6572.65
GaLoreBackbone=Qwen2.5-7B2026.0568.3282.8766.0876.172.55
LoRABackbone=Qwen2.5-7B2026.0567.6982.9166.2975.7272.41
LoRA + MoLSBackbone=LLaMA3.2-3B2026.0547.3465.7749.6162.4155.62
LoRABackbone=LLaMA3.2-3B2026.0547.2263.4149.5661.5754.86
GaLoreBackbone=LLaMA3.2-3B2026.0546.6265.5848.3763.3355.22
FullBackbone=LLaMA3.2-3B2026.0546.5565.8149.1663.1755.48
APOLLOBackbone=LLaMA3.2-3B2026.0546.1665.9149.0362.855.29
FullBackbone=Gemma3-1B2026.0527.6326.8425.1225.1426.04
LoRA + MoLSBackbone=Gemma3-1B2026.0527.5126.9625.6725.4926.31
APOLLOBackbone=Gemma3-1B2026.0527.4727.1425.6125.7626.38
GaLoreBackbone=Gemma3-1B2026.0527.426.7124.825.6925.99
LoRABackbone=Gemma3-1B2026.0526.6125.9324.7825.0225.48