List your company

Mathematical Reasoning on AIME (acc, len)

35.69AIME AccuracySA-GRPO
-1.09488.455118.00527.5549Sep 5, 2025

Evaluation results

MethodLinks
SA-GRPOBase Model=Qwen3-4B, alpha=0.22025.0935.691,788
GRPOBase Model=Qwen3-4B2025.0930.112,113
GRPO-ERBase Model=Qwen3-4B, alpha=0.052025.0928.941,022
GRPO-O1Base Model=Qwen3-4B, alpha=0.052025.0928.621,016
SA-PPOBase Model=Qwen3-4B2025.0928.191,397
PPOBase Model=Qwen3-4B2025.0926.92,680
Qwen3-4BBase Model=Qwen3-4B2025.0920.151,281
Gemma3-4BBackbone=Gemma3-4B, Strategy=Base2025.0918.761,843
GRPO-O1Backbone=Gemma3-4B2025.0918.541,814
SA-GRPOBackbone=Gemma3-4B2025.0918.112,379
GRPOBackbone=Gemma3-4B2025.0918.012,357
GRPO-ERBackbone=Gemma3-4B2025.0917.362,268
GRPOBase Model=Qwen3-1.7B2025.0913.931,278
GRPO-O1Base Model=Qwen3-1.7B, alpha=0.052025.0913.61910
SA-GRPOBase Model=Qwen3-1.7B, alpha=0.22025.0913.5992
GRPO-ERBase Model=Qwen3-1.7B, alpha=0.052025.0912.221,232
SA-PPOBase Model=Qwen3-1.7B2025.0910.611,015
Qwen3-1.7BBase Model=Qwen3-1.7B2025.099.971,539
PPOBase Model=Qwen3-1.7B2025.099.111,718
PPOBase Model=Phi-3.5-mini2025.093.85926
SA-GRPOBase Model=Phi-3.5-mini, alpha=0.22025.093.22535
GRPOBase Model=Phi-3.5-mini2025.093.21672
SA-PPOBase Model=Phi-3.5-mini2025.092.79501
GRPO-O1Base Model=Phi-3.5-mini, alpha=0.052025.092.57583
GRPO-ERBase Model=Phi-3.5-mini, alpha=0.052025.092.47606
Phi-3.5-miniBase Model=Phi-3.5-mini2025.092.35943
GRPO-O1Base Model=Gemma3-1B, alpha=0.052025.092.352,449
Gemma3-1BBase Model=Gemma3-1B2025.092.252,279
GRPOBase Model=Gemma3-1B2025.092.252,613
SA-GRPOBase Model=Gemma3-1B, alpha=0.22025.092.141,509
GRPO-ERBase Model=Gemma3-1B, alpha=0.052025.092.041,696
PPOBase Model=Gemma3-1B2025.091.62,422
SA-PPOBase Model=Gemma3-1B2025.091.51,683
SA-GRPOBackbone=Qwen2-2B-VL2025.091.2957
GRPOBackbone=Qwen2-2B-VL2025.090.86561
Qwen2-2B-VLBackbone=Qwen2-2B-VL, Strategy=Base2025.090.542,404
GRPO-ERBackbone=Qwen2-2B-VL2025.090.43383
GRPO-O1Backbone=Qwen2-2B-VL2025.090.3217