List your company
ResearchFormerly WizWand

Mathematical Reasoning on GSM8K (Pass Rate)

96Pass RateGEM
57.72867.66477.687.536Feb 7, 2026
Updated 8mo ago

Evaluation results

MethodLinks
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0296
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0296
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0295.3
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0295.2
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0294.9
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0294.5
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.0287.7
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.0287.7
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0287.4
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.0286.8
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0285.1
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0284
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0283.5
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0281
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0278.6
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0269.9
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.0261.2
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.0260.7
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0260.1
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.0259.2