List your company
ResearchFormerly WizWand

Mathematical Reasoning on MathQA OOD (test)

63AccuracyRKU
28.451237.420646.3955.3594May 9, 2026
Updated 5mo ago

Evaluation results

MethodLinks
RKUSparsity=40%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0563
RKUSparsity=30%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0561.67
Wanda-StructSparsity=50%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0560.33
RKUSparsity=50%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0560
Wanda-StructSparsity=40%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0559.33
Taylor-FOSparsity=50%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0556.33
Taylor-FOSparsity=40%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0554.33
Taylor-FOSparsity=30%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0553.33
Wanda-StructSparsity=30%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0551.67
AGFSparsity=30%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0542.58
RKUSparsity=30%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0541.88
Taylor-FOSparsity=30%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0541.54
Wanda-StructSparsity=30%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0540.2
AGFSparsity=30%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0539.13
AGFSparsity=40%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0538.46
RKUSparsity=40%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0537.45
Wanda-StructSparsity=40%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0535.71
Taylor-FOSparsity=40%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0535.38
AGFSparsity=50%, Training Source=GSM8K, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0535.04
AGFSparsity=40%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0534.81
RKUSparsity=50%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0532.46
Taylor-FOSparsity=50%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0530.79
Wanda-StructSparsity=50%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0530.59
AGFSparsity=50%, Training Source=AQuA, Backbone=Qwen-2.5-7B, Fine-tuning steps=150, PEFT Method=LoRA2026.0529.78