List your company

Reasoning on BBH, MMLU, ARC-C, and ThmQA (test)

64.66BBHTeacher
1.37617.805534.23550.6645May 25, 2026

Evaluation results

MethodLinks
TeacherModel=Qwen2.5-7B2026.0564.6678.2289.932.4766.31
TeacherTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0557.7270.983.5818.157.58
TeacherTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.0552.5766.1781.2326.1256.52
HybKDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0546.5369.0581.2323.8255.16
HybKDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0546.5369.0581.2323.8255.16
Hybrid KDSetting=Qwen2.5-32B→3B2026.0545.766.8802454.1
JS divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0545.564.6878.8522.2752.83
JS divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0545.564.6878.8522.2752.83
α-β divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0545.1264.9579.8122.9453.21
α-β divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0545.1264.9579.8122.9453.21
Adaptive KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0544.7164.6979.2322.2552.72
Adaptive KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0544.7164.6979.2322.2552.72
Reverse KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0544.0765.6777.6824.252.9
Reverse KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0544.0765.6777.6824.252.9
Soft KDSetting=Qwen2.5-32B→3B2026.054465.678.122.952.6
Forward KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.6564.4578.3323.0251.87
Hard KDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.5265.7678.7523.7552.45
Skew FKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.3964.6777.7523.7751.89
Skew FKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.3964.6777.7523.7751.89
Skew RKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.2263.9576.9123.6751.44
Skew RKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0541.2263.9576.9123.6751.44
Total VariationTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0540.564.5278.1122.8351.49
Total VariationTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0540.564.5278.1122.8351.49
Hard KDSetting=Qwen2.5-32B→3B2026.0534.467.279.523.151
Skew RKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0528.1645.0347.5111.3733.02
Skew RKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0528.1645.0347.5111.3733.02
HybKDTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0527.4435.6437.01526.27
Total VariationTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.7444.3546.7610.232.01
Total VariationTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.7444.3546.7610.232.01
HybKDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.5849.0851.6910.534.46
HybKDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.5849.0851.6910.534.46
α-β divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.1842.5946.711.0731.63
α-β divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.1842.5946.711.0731.63
Adaptive KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.0441.8545.8911.3331.28
Adaptive KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0526.0441.8545.8911.3331.28
Skew FKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0525.8744.1247.1210.6531.94
Skew FKLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0525.8744.1247.1210.6531.94
Skew RKLTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0525.243332.244.2723.69
Adaptive KLTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0525.1833.8633.744.8524.41
JS divergenceTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0525.1634.0834.425.724.84
α-β divergenceTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0525.0734.7834.95.2725.01
Skew FKLTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0525.0531.3932.344.7323.37
Reverse KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0524.9144.7247.441132.02
Reverse KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0524.9144.7247.441132.02
JS divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0524.5543.3144.7811.8231.12
JS divergenceTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0524.5543.3144.7811.8231.12
Forward KLTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0524.4243.1946.8410.3331.2
Reverse KLTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0523.6931.6332.083.622.75
Total VariationTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0523.5527.4128.992.6820.66
Student (No Distill)Teacher Model=Qwen2.5-7B, Student Model=Qwen2.5-3B2026.0522.3464.6178.412.2244.39
Forward KLTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0522.0733.1333.414.3723.25
Hard KDTeacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0515.2922.5423.983.8816.42
Student (No Distill)Teacher-Student Model Pair=Llama3.1-8B → Llama3.2-1B2026.0514.0119.7821.572.2214.4
Hard KDTeacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.0510.2736.8340.5811.3524.76
Reverse KLTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.059.5224.7727.581.2615.78
HybKDTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.057.7825.9426.075.2416.26
α-β divergenceTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.057.2625.4926.074.3415.79
JS divergenceTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.057.1425.2726.034.5715.75
Skew FKLTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.057.0125.0826.145.0115.81
Student (No Distill)Teacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.056.4715.9415.70.959.76
Total VariationTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.056.2925.0725.831.2914.62
Forward KLTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.055.9124.05254.2514.8
Skew RKLTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.055.8325.1725.933.3815.08
Adaptive KLTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.055.4724.7624.074.2314.63
Hard KDTeacher-Student Model Pair=Gemma3-4B → Gemma3-1B2026.055.2818.1118.093.1511.16
Student (No Distill)Teacher Model=Qwen2.5-7B, Student Model=Qwen2.5-0.5B2026.053.8144.6743.727.7824.99