List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC23 (Accuracy Statistics)

70.4Mean AccuracyFlowTracer
20.58433.51746.4559.383Jun 8, 2026Jun 9, 2026
Updated 4mo ago

Evaluation results

MethodLinks
FlowTracerBackbone=Qwen3-8B-Base, Context Length=8K2026.0670.4——
RandomBackbone=Qwen3-8B-Base, Context Length=8K2026.0667.2——
GRPOBackbone=Qwen3-8B-Base, Context Length=8K2026.0666.3——
AttentionBackbone=Qwen3-8B-Base, Context Length=8K2026.0666.1——
GradientBackbone=Qwen3-8B-Base, Context Length=8K2026.0665.7——
FlowTracerBackbone=Qwen3-8B-Base, Context Length=1K2026.0665.6——
CorrelationBackbone=Qwen3-8B-Base, Context Length=8K2026.0664.4——
High-entropyBackbone=Qwen3-8B-Base, Context Length=8K2026.0663.7——
AttentionBackbone=Qwen3-8B-Base, Context Length=1K2026.0662.9——
FlowTracerBackbone=Qwen3-4B-Base, Context Length=8K2026.0662.4——
High-entropyBackbone=Qwen3-8B-Base, Context Length=1K2026.0661.5——
RandomBackbone=Qwen3-8B-Base, Context Length=1K2026.0660.1——
GradientBackbone=Qwen3-8B-Base, Context Length=1K2026.0660——
AttentionBackbone=Qwen3-4B-Base, Context Length=8K2026.0659.3——
GRPOBackbone=Qwen3-8B-Base, Context Length=1K2026.0659.1——
FlowTracerBackbone=Qwen3-4B-Base, Context Length=1K2026.0659——
CorrelationBackbone=Qwen3-8B-Base, Context Length=1K2026.0658.8——
AttentionBackbone=Qwen3-4B-Base, Context Length=1K2026.0658.4——
High-entropyBackbone=Qwen3-4B-Base, Context Length=8K2026.0657.8——
GradientBackbone=Qwen3-4B-Base, Context Length=8K2026.0657.8——
CorrelationBackbone=Qwen3-4B-Base, Context Length=8K2026.0657.8——
GRPOBackbone=Qwen3-4B-Base, Context Length=8K2026.0657.6——
RandomBackbone=Qwen3-4B-Base, Context Length=8K2026.0657.1——
High-entropyBackbone=Qwen3-4B-Base, Context Length=1K2026.0655.8——
RandomBackbone=Qwen3-4B-Base, Context Length=1K2026.0655.2——
CorrelationBackbone=Qwen3-4B-Base, Context Length=1K2026.0655.2——
GRPOBackbone=Qwen3-4B-Base, Context Length=1K2026.0655.1——
GradientBackbone=Qwen3-4B-Base, Context Length=1K2026.0653.2——
RePO_detModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06457.7135.43
KTOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0643.56.7535.11
TDPOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06423.2637.95
RPOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0641.54.1836.31
RePOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06388.5527.38
DPOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0635.55.728.42
IPOModel (Backbone)=Qwen3-4B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0635528.79
RePOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06334.8127.03
RePO_detModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0630.53.2626.45
KTOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06303.0626.2
TDPOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06268.415.57
DPOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06253.9520.09
IPOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.06243.3519.84
RPOModel (Backbone)=Qwen3-1.7B, Decoding Strategy=sampling-based inference, Temperature=0.7, Top-p=0.952026.0622.53.0618.7