List your company

Mathematical Reasoning on NuminaMath

60.9AccuracyQwen2.5-Math-72B-Instruct
1.3616.817532.27547.7325Mar 4, 2025Apr 3, 2025May 4, 2025Jun 4, 2025Jul 5, 2025Aug 5, 2025Sep 5, 2025

Evaluation results

MethodLinks
Qwen2.5-Math-72B-InstructEvaluating Model=Qwen2.5-Math-72B-Instruct2025.0360.9—
SA-GRPOBase Model=Qwen3-4B, alpha=0.22025.0957.71,358
GRPOBase Model=Qwen3-4B2025.0953.791,485
GRPO-O1Base Model=Qwen3-4B, alpha=0.052025.0951.06866
PPOBase Model=Qwen3-4B2025.0950.81,995
GRPO-ERBase Model=Qwen3-4B, alpha=0.052025.0950.47854
SA-PPOBase Model=Qwen3-4B2025.0949.091,133
GRPO-ERBackbone=Gemma3-4B2025.0941.941,817
GRPOBackbone=Gemma3-4B2025.0941.92,012
GRPO-O1Backbone=Gemma3-4B2025.0941.681,435
SA-GRPOBackbone=Gemma3-4B2025.0940.481,920
Qwen3-4BBase Model=Qwen3-4B2025.0938.81,142
SA-GRPOBase Model=Qwen3-1.7B, alpha=0.22025.0936.03841
Gemma3-4BBackbone=Gemma3-4B, Strategy=Base2025.0935.811,370
GRPO-ERBase Model=Qwen3-1.7B, alpha=0.052025.0934.35980
GRPOBase Model=Qwen3-1.7B2025.0933.441,091
GRPO-O1Base Model=Qwen3-1.7B, alpha=0.052025.0933.15775
SA-PPOBase Model=Qwen3-1.7B2025.0932.05866
PPOBase Model=Qwen3-1.7B2025.0927.021,358
Qwen3-1.7BBase Model=Qwen3-1.7B2025.0924.581,242
SA-PPOBase Model=Phi-3.5-mini2025.0917.43586
SA-GRPOBase Model=Gemma3-1B, alpha=0.22025.0916.71,302
GRPOBase Model=Gemma3-1B2025.0916.592,264
GRPO-ERBase Model=Gemma3-1B, alpha=0.052025.0916.411,545
GRPOBase Model=Phi-3.5-mini2025.0916.37582
GRPO-O1Base Model=Gemma3-1B, alpha=0.052025.0916.161,871
SA-GRPOBase Model=Phi-3.5-mini, alpha=0.22025.0916.08455
PPOBase Model=Gemma3-1B2025.0915.571,904
SA-PPOBase Model=Gemma3-1B2025.0914.221,523
GRPO-O1Base Model=Phi-3.5-mini, alpha=0.052025.0914.08509
GRPO-ERBase Model=Phi-3.5-mini, alpha=0.052025.0912.55511
Gemma3-1BBase Model=Gemma3-1B2025.0912.41,414
Phi-3.5-miniBase Model=Phi-3.5-mini2025.0911.561,492
PPOBase Model=Phi-3.5-mini2025.0911.451,707
GRPO-O1Backbone=Qwen2-2B-VL2025.099.9922
SA-GRPOBackbone=Qwen2-2B-VL2025.099.5256
GRPOBackbone=Qwen2-2B-VL2025.099.19512
GRPO-ERBackbone=Qwen2-2B-VL2025.098.32330
Qwen2-2B-VLBackbone=Qwen2-2B-VL, Strategy=Base2025.093.652,411
DeepSeek-R1-Distill-Qwen-7BEvaluating Model=DeepSeek-R1-Distill-Qwen-7B2025.03—3,039