List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC23 (Pass Rate Avg@8)

67.8Pass Rate (Avg@8)CrossEntropy
823.52539.0554.575Feb 7, 2026
Updated 8mo ago

Evaluation results

MethodLinks
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0267.8
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0267.2
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0266.2
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0263.7
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0261.6
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0258.1
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.0246.9
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.0242.2
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0240
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0238.4
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.0235.3
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0233.4
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0231.5
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0227.3
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0226.6
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.0217.2
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0216.6
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.0216.2
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0212.8
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.0210.3