List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC12

93.97Mean@4Sparsity setup (0.86) Rollout
10.87432.44754.0275.593Feb 5, 2026Feb 25, 2026Mar 17, 2026Apr 7, 2026Apr 27, 2026May 17, 2026Jun 7, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Sparsity setup (0.86) RolloutModel Backbone=Qwen3-4B, Sparsity Level=0.862026.0693.97
Dense Rollout Dense TrainingModel Backbone=Qwen3-14B2026.0693.96
Dense Rollout Dense TrainingModel Backbone=Qwen3-4B2026.0693.67
Sparsity setup (0.86) RolloutModel Backbone=Qwen3-14B, Sparsity Level=0.862026.0693.67
Sparsity setup (0.92) RolloutModel Backbone=Qwen3-4B, Sparsity Level=0.922026.0693.37
Dense Rollout Dense TrainingModel Backbone=Qwen3-8B2026.0693.37
Sparsity setup (0.92) RolloutModel Backbone=Qwen3-8B, Sparsity Level=0.922026.0692.7
Sparsity setup (0.86) RolloutModel Backbone=Qwen3-8B, Sparsity Level=0.862026.0692.16
Sparse with LoRA Distillation (DistillSparse)Model Backbone=Qwen3-1.7B2026.0680.72
Dense Rollout Dense TrainingModel Backbone=Qwen3-1.7B2026.0673.33
Sparsity setup (0.92) RolloutModel Backbone=Qwen3-1.7B, Sparsity Level=0.922026.0672.22
Sparsity setup (0.86) RolloutModel Backbone=Qwen3-1.7B, Sparsity Level=0.862026.0671.66
JackpotTraining Scenario=Qwen3-1.7B -> Qwen3-8B (15k, 64, DeepScaleR)2026.0254.44
JackpotTraining Scenario=Qwen3-1.7B -> Qwen3-4B (20k, 64, DeepScaleR)2026.0253.88
Q3-8BTraining Scenario=Qwen3-1.7B -> Qwen3-8B (15k, 64, DeepScaleR), Configuration=On-policy2026.0253.33
Q3-4BTraining Scenario=Qwen3-1.7B -> Qwen3-4B (20k, 64, DeepScaleR), Configuration=On-policy2026.0251.66
TIS + Reverse KLTraining Scenario=Qwen3-1.7B -> Qwen3-8B (15k, 64, DeepScaleR)2026.0237.22
TIS + Reverse KLTraining Scenario=Qwen3-1.7B -> Qwen3-4B (20k, 64, DeepScaleR)2026.0232.77
JackpotTraining Scenario=Qwen2.5-1.5B -> Qwen2.5-3B (14k, 64, MATH-8K)2026.0227.78
Qwen2.5-3BTraining Scenario=Qwen2.5-1.5B -> Qwen2.5-3B (14k, 64, MATH-8K), Configuration=On-policy2026.0226.11
Q3-1.7BTraining Scenario=Qwen3-1.7B -> Qwen3-4B (20k, 64, DeepScaleR), Configuration=On-policy2026.0225
TIS + Reverse KLTraining Scenario=Qwen2.5-1.5B -> Qwen2.5-3B (14k, 64, MATH-8K)2026.0224.44
Only Reverse KLTraining Scenario=Qwen2.5-1.5B -> Qwen2.5-3B (14k, 64, MATH-8K)2026.0214.07