List your company
ResearchFormerly WizWand

Multi-domain Language Model Evaluation on Knowledge, Math, Code, Reasoning, and Biomedical Domains

69.2Average AccuracyCODA
54.84858.57462.366.026Jul 3, 2026
Updated 3mo ago

Evaluation results

MethodLinks
CODABackbone=Gemma-2-9B, Controller Budget=50%2026.0769.2
GradNormBackbone=Gemma-2-9B, Controller Budget=50%2026.0768.3
CODABackbone=Qwen-2.5-7B, Controller Budget=50%2026.0768
Full dataBackbone=Gemma-2-9B2026.0767.2
GradNormBackbone=Qwen-2.5-7B, Controller Budget=50%2026.0766.8
Full dataBackbone=Qwen-2.5-7B2026.0766.2
CODABackbone=Qwen-2.5-3B, Controller Budget=50%2026.0763.6
GradNormBackbone=Qwen-2.5-3B, Controller Budget=50%2026.0762.8
Full dataBackbone=Qwen-2.5-3B2026.0761.5
CODABackbone=Mistral-7B, Controller Budget=50%2026.0760.3
GradNormBackbone=Mistral-7B, Controller Budget=50%2026.0759.4
Full dataBackbone=Mistral-7B2026.0758.2
CODABackbone=LLaMA-3.1-8B, Controller Budget=50%2026.0757.2
GradNormBackbone=LLaMA-3.1-8B, Controller Budget=50%2026.0756
Full dataBackbone=LLaMA-3.1-8B2026.0755.4