Mathematical Reasoning on GSM8K (Accuracy, Spread)
84AccuracyDense
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DenseBackbone=Qwen3-4B-Instruct, Sparsity=Dense2026.06 | 84 | — | |
| DenseBackbone=Qwen2.5-7B-Instruct, Sparsity=Dense2026.06 | 81.2 | — | |
| DenseBackbone=Llama-3.1-8B-Instruct, Sparsity=Dense2026.06 | 79.2 | — | |
| DenseBackbone=Llama-3.2-3B-Instruct, Sparsity=Dense2026.06 | 67.7 | — | |
| RM-LSBackbone=Llama-3.1-8B-Instruct, Sparsity=12.5%2026.06 | 65.5 | 9.2 | |
| RM-Cos.Backbone=Llama-3.1-8B-Instruct, Sparsity=12.5%2026.06 | 65.3 | 9.2 | |
| SL-LayerBackbone=Llama-3.1-8B-Instruct, Sparsity=12.5%2026.06 | 62.3 | 9.2 | |
| SL-FFNBackbone=Llama-3.1-8B-Instruct, Sparsity=12.5%2026.06 | 62.1 | 9.2 | |
| SUBFITBackbone=Qwen3-4B-Instruct, Sparsity=12.5%2026.06 | 60.7 | 48 | |
| SUBFITBackbone=Llama-3.1-8B-Instruct, Sparsity=12.5%2026.06 | 56.3 | 9.2 | |
| RM-Cos.Backbone=Llama-3.1-8B-Instruct, Sparsity=20%2026.06 | 52.8 | 22.2 | |
| DenseBackbone=DeepSeek-7B-chat, Sparsity=Dense2026.06 | 51 | — | |
| SL-LayerBackbone=Llama-3.2-3B-Instruct, Sparsity=12.5%2026.06 | 47.5 | 45.3 | |
| RM-Cos.Backbone=Llama-3.2-3B-Instruct, Sparsity=12.5%2026.06 | 40.9 | 45.3 | |
| RM-LSBackbone=Llama-3.2-3B-Instruct, Sparsity=12.5%2026.06 | 40.8 | 45.3 | |
| RM-LSBackbone=Llama-3.1-8B-Instruct, Sparsity=20%2026.06 | 40.6 | 22.2 | |
| RM-Cos.Backbone=Qwen3-4B-Instruct, Sparsity=12.5%2026.06 | 39.7 | 48 | |
| SL-FFNBackbone=Llama-3.1-8B-Instruct, Sparsity=20%2026.06 | 39.7 | 22.2 | |
| SL-FFNBackbone=Qwen3-4B-Instruct, Sparsity=12.5%2026.06 | 38.5 | 48 | |
| SL-FFNBackbone=Llama-3.2-3B-Instruct, Sparsity=12.5%2026.06 | 32.1 | 45.3 | |
| RM-LSBackbone=Qwen2.5-7B-Instruct, Sparsity=12.5%2026.06 | 31.9 | 24.6 | |
| SL-LayerBackbone=Qwen3-4B-Instruct, Sparsity=12.5%2026.06 | 31.7 | 48 | |
| SL-LayerBackbone=Llama-3.1-8B-Instruct, Sparsity=20%2026.06 | 31.6 | 22.2 | |
| SUBFITBackbone=Llama-3.1-8B-Instruct, Sparsity=20%2026.06 | 30.6 | 22.2 | |
| RM-Cos.Backbone=Qwen2.5-7B-Instruct, Sparsity=12.5%2026.06 | 30 | 24.6 | |
| SUBFITBackbone=DeepSeek-7B-chat, Sparsity=12.5%2026.06 | 25.4 | 3.4 | |
| SL-FFNBackbone=DeepSeek-7B-chat, Sparsity=12.5%2026.06 | 24.6 | 3.4 | |
| RM-LSBackbone=DeepSeek-7B-chat, Sparsity=12.5%2026.06 | 24.2 | 3.4 | |
| SUBFITBackbone=Qwen2.5-7B-Instruct, Sparsity=12.5%2026.06 | 24.1 | 24.6 | |
| SL-LayerBackbone=DeepSeek-7B-chat, Sparsity=12.5%2026.06 | 23.1 | 3.4 | |
| SUBFITBackbone=Qwen3-4B-Instruct, Sparsity=20%2026.06 | 22 | 19.7 | |
| RM-Cos.Backbone=DeepSeek-7B-chat, Sparsity=12.5%2026.06 | 22 | 3.4 | |
| RM-Cos.Backbone=Llama-3.1-8B-Instruct, Sparsity=25%2026.06 | 16.7 | 13.1 | |
| SL-LayerBackbone=Qwen2.5-7B-Instruct, Sparsity=12.5%2026.06 | 16.1 | 24.6 | |
| SUBFITBackbone=DeepSeek-7B-chat, Sparsity=20%2026.06 | 12.8 | 9.5 | |
| RM-LSBackbone=Qwen3-4B-Instruct, Sparsity=12.5%2026.06 | 12.7 | 48 | |
| SL-LayerBackbone=Llama-3.2-3B-Instruct, Sparsity=20%2026.06 | 11.6 | 9 | |
| RM-LSBackbone=Llama-3.1-8B-Instruct, Sparsity=25%2026.06 | 10.5 | 13.1 | |
| RM-LSBackbone=DeepSeek-7B-chat, Sparsity=20%2026.06 | 9.7 | 9.5 | |
| SL-LayerBackbone=DeepSeek-7B-chat, Sparsity=20%2026.06 | 9.5 | 9.5 | |
| RM-LSBackbone=Llama-3.2-3B-Instruct, Sparsity=20%2026.06 | 9.1 | 9 | |
| RM-Cos.Backbone=Llama-3.2-3B-Instruct, Sparsity=20%2026.06 | 8.5 | 9 | |
| SUBFITBackbone=Llama-3.1-8B-Instruct, Sparsity=25%2026.06 | 8.3 | 13.1 | |
| SL-FFNBackbone=Qwen2.5-7B-Instruct, Sparsity=12.5%2026.06 | 7.3 | 24.6 | |
| SUBFITBackbone=Qwen2.5-7B-Instruct, Sparsity=20%2026.06 | 7.1 | 5.1 | |
| RM-LSBackbone=Qwen2.5-7B-Instruct, Sparsity=20%2026.06 | 6.3 | 5.1 | |
| RM-Cos.Backbone=Qwen3-4B-Instruct, Sparsity=20%2026.06 | 5.6 | 19.7 | |
| RM-Cos.Backbone=DeepSeek-7B-chat, Sparsity=20%2026.06 | 5.2 | 9.5 | |
| RM-LSBackbone=Llama-3.2-3B-Instruct, Sparsity=25%2026.06 | 5 | 3.9 | |
| RM-Cos.Backbone=Qwen2.5-7B-Instruct, Sparsity=20%2026.06 | 4.5 | 5.1 | |
| SL-LayerBackbone=Qwen2.5-7B-Instruct, Sparsity=20%2026.06 | 4.1 | 5.1 | |
| SL-FFNBackbone=Qwen3-4B-Instruct, Sparsity=20%2026.06 | 4 | 19.7 | |
| SL-FFNBackbone=Llama-3.1-8B-Instruct, Sparsity=25%2026.06 | 4 | 13.1 | |
| SL-LayerBackbone=Llama-3.2-3B-Instruct, Sparsity=25%2026.06 | 3.8 | 3.9 | |
| SUBFITBackbone=Qwen2.5-7B-Instruct, Sparsity=25%2026.06 | 3.6 | 1.8 | |
| SUBFITBackbone=DeepSeek-7B-chat, Sparsity=25%2026.06 | 3.6 | 2 | |
| SL-LayerBackbone=Llama-3.1-8B-Instruct, Sparsity=25%2026.06 | 3.6 | 13.1 | |
| SL-FFNBackbone=Llama-3.2-3B-Instruct, Sparsity=25%2026.06 | 3.5 | 3.9 | |
| SL-FFNBackbone=DeepSeek-7B-chat, Sparsity=20%2026.06 | 3.3 | 9.5 | |
| SUBFITBackbone=DeepSeek-7B-chat, Sparsity=30%2026.06 | 3.3 | 3.1 | |
| RM-Cos.Backbone=Llama-3.2-3B-Instruct, Sparsity=25%2026.06 | 3.2 | 3.9 | |
| RM-LSBackbone=Llama-3.2-3B-Instruct, Sparsity=30%2026.06 | 2.9 | 1.1 | |
| SUBFITBackbone=Qwen3-4B-Instruct, Sparsity=25%2026.06 | 2.9 | 2.4 | |
| SUBFITBackbone=Llama-3.2-3B-Instruct, Sparsity=20%2026.06 | 2.7 | 9 | |
| RM-LSBackbone=Llama-3.1-8B-Instruct, Sparsity=30%2026.06 | 2.7 | 1.2 | |
| SL-FFNBackbone=Llama-3.2-3B-Instruct, Sparsity=20%2026.06 | 2.6 | 9 | |
| RM-Cos.Backbone=Qwen2.5-7B-Instruct, Sparsity=25%2026.06 | 2.6 | 1.8 | |
| RM-Cos.Backbone=Qwen2.5-7B-Instruct, Sparsity=30%2026.06 | 2.6 | 1.2 | |
| RM-Cos.Backbone=Llama-3.1-8B-Instruct, Sparsity=30%2026.06 | 2.6 | 1.2 | |
| SL-LayerBackbone=Qwen3-4B-Instruct, Sparsity=20%2026.06 | 2.5 | 19.7 | |
| RM-LSBackbone=Qwen2.5-7B-Instruct, Sparsity=30%2026.06 | 2.5 | 1.2 | |
| SL-LayerBackbone=Llama-3.2-3B-Instruct, Sparsity=30%2026.06 | 2.4 | 1.1 | |
| RM-Cos.Backbone=Llama-3.2-3B-Instruct, Sparsity=37.5%2026.06 | 2.4 | 0.6 | |
| RM-LSBackbone=Qwen2.5-7B-Instruct, Sparsity=25%2026.06 | 2.4 | 1.8 | |
| SUBFITBackbone=Llama-3.2-3B-Instruct, Sparsity=12.5%2026.06 | 2.3 | 45.3 | |
| RM-LSBackbone=Qwen3-4B-Instruct, Sparsity=20%2026.06 | 2.3 | 19.7 | |
| RM-Cos.Backbone=Qwen3-4B-Instruct, Sparsity=30%2026.06 | 2.3 | 2.3 | |
| RM-Cos.Backbone=DeepSeek-7B-chat, Sparsity=25%2026.06 | 2.3 | 2 | |
| SUBFITBackbone=Qwen3-4B-Instruct, Sparsity=30%2026.06 | 2.2 | 2.3 | |
| RM-Cos.Backbone=Qwen2.5-7B-Instruct, Sparsity=37.5%2026.06 | 2.2 | 0.9 | |
| RM-Cos.Backbone=DeepSeek-7B-chat, Sparsity=37.5%2026.06 | 2.2 | 2.2 | |
| RM-Cos.Backbone=Llama-3.2-3B-Instruct, Sparsity=30%2026.06 | 2.1 | 1.1 | |
| SUBFITBackbone=Llama-3.2-3B-Instruct, Sparsity=37.5%2026.06 | 2.1 | 0.6 | |
| SL-FFNBackbone=Llama-3.2-3B-Instruct, Sparsity=30%2026.06 | 2 | 1.1 | |
| RM-LSBackbone=Llama-3.2-3B-Instruct, Sparsity=37.5%2026.06 | 2 | 0.6 | |
| SL-FFNBackbone=Qwen3-4B-Instruct, Sparsity=25%2026.06 | 2 | 2.4 | |
| RM-LSBackbone=Qwen3-4B-Instruct, Sparsity=25%2026.06 | 2 | 2.4 | |
| SL-FFNBackbone=Qwen2.5-7B-Instruct, Sparsity=20%2026.06 | 2 | 5.1 | |
| SL-LayerBackbone=DeepSeek-7B-chat, Sparsity=25%2026.06 | 2 | 2 | |
| RM-LSBackbone=DeepSeek-7B-chat, Sparsity=25%2026.06 | 2 | 2 | |
| RM-Cos.Backbone=Llama-3.1-8B-Instruct, Sparsity=37.5%2026.06 | 2 | 0.5 | |
| SUBFITBackbone=Llama-3.1-8B-Instruct, Sparsity=37.5%2026.06 | 2 | 0.5 | |
| SL-LayerBackbone=Llama-3.2-3B-Instruct, Sparsity=37.5%2026.06 | 1.9 | 0.6 | |
| SUBFITBackbone=Llama-3.2-3B-Instruct, Sparsity=30%2026.06 | 1.8 | 1.1 | |
| RM-LSBackbone=Qwen3-4B-Instruct, Sparsity=30%2026.06 | 1.8 | 2.3 | |
| SL-FFNBackbone=Qwen2.5-7B-Instruct, Sparsity=25%2026.06 | 1.8 | 1.8 | |
| SL-FFNBackbone=Qwen2.5-7B-Instruct, Sparsity=37.5%2026.06 | 1.8 | 0.9 | |
| SUBFITBackbone=Llama-3.1-8B-Instruct, Sparsity=30%2026.06 | 1.8 | 1.2 | |
| SL-FFNBackbone=Llama-3.2-3B-Instruct, Sparsity=37.5%2026.06 | 1.7 | 0.6 | |
| RM-Cos.Backbone=Qwen3-4B-Instruct, Sparsity=25%2026.06 | 1.7 | 2.4 |