List your company

Mathematical Reasoning on GSM8k 2021 (test)

100Accuracyentropy-guided decoding
64.95274.05183.1592.249Mar 10, 2026Mar 15, 2026Mar 21, 2026Mar 26, 2026Apr 1, 2026Apr 6, 2026Apr 12, 2026

Evaluation results

MethodLinks
entropy-guided decodingModel=GPT4o2026.0310096.31.31998.799.642.2
entropy-guided decodingModel=Llama3.1-8B2026.0399.684.21.42399.399.730.2
SFT No-TestModel variant=SFT No-Test2026.0480.7——————
Merged (α=0.2)Model variant=Merged (α=0.2), Interpolation Weight=0.22026.0468.3——————
Qwen3-4B-Instruct-2507Model variant=Base Model2026.0466.3——————