List your company
ResearchFormerly WizWand

Mathematical Problem Solving on AMC 2023 (Accuracy avg@k)

97.81Accuracy (avg@k)TRACE (RKL on Ey)
23.335642.670362.00581.3397May 11, 2026May 17, 2026May 24, 2026May 31, 2026Jun 7, 2026Jun 14, 2026Jun 21, 2026
Updated 4mo ago

Evaluation results

MethodLinks
TRACE (RKL on Ey)Backbone=Qwen3-8B, Symmetric Think evaluation=true2026.0597.81
TRACE (FKL on Ky)Backbone=Qwen3-8B, Symmetric Think evaluation=true2026.0597.66
RLSDBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.0597.03
GRPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.0596.56
BaseBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.0595.94
SRPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.0593.12
SDPOBackbone=Qwen3-8B, Symmetric Think evaluation=true2026.0591.41
TRACE (RKL on Ey)Backbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0586.82
BaseBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0586.56
RLSDBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0584.38
TRACE (FKL on Ky)Backbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0583.38
GRPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0581.56
SRPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0562.81
ACPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0654.37
SDPOBackbone=Qwen3-1.7B, Symmetric Think evaluation=true2026.0551.25
DAPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0651.2
ACPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0651.2
High-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0650.75
CISPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0649.25
Low-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0648.95
DAPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0646.84
High-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0646.39
CISPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0646.08
AR-LoptiRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0644.88
Low-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0640.81
AR-LoptiRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0634.94
Base ModelRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0626.2