ResearchBenchmarksMathematical Reasoning on GSM8K (ACC, 95% CI, S-ACC)Follow82.03ACCSD-E275.176476.955778.73580.5143Jan 25, 2026Evaluation ResultsMethodMethodLinksACCACC 95% CIS-ACCSD-E2Backbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Inst., #STR=9.78, #TOK=291.422026.0182.03—97.2GRPO-CFEEBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Inst., #STR=5.7, #TOK=278.542026.0180.52—92.3GRPO-CFLBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Inst., #TOK=265.722026.0176.8——SD-E2Backbone=Llama-3.1-8B-...Backbone=Llama-3.1-8B-Inst., #STR=8.58, #TOK=287.512026.0175.44—95