List your company

Mathematical Reasoning on AMC 2023 (pass@1 metrics)

86.3Pass@1 Success Rate (Last)SDPG-URKL
70.80474.82778.8582.873Jun 2, 2026

Evaluation results

MethodLinks
SDPG-URKLBackbone=Qwen3-4B, Sampling Strategy=mean@32, Training Steps=4002026.0686.386.3
SDPG-UFKLBackbone=Qwen3-4B, Sampling Strategy=mean@32, Training Steps=4002026.0685.887
RLSDBackbone=Qwen3-4B, Sampling Strategy=mean@32, Training Steps=4002026.0681.381.3
GRPOBackbone=Qwen3-4B, Sampling Strategy=mean@32, Training Steps=4002026.0671.473.9