Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for 13B FKD post-training performance.
来源：source_paper:Table 2, Page 9
报告指标与观测值
downstream_accuracy
t2_13b_fkd_gsm8k
论文报告 72.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_gsm_s
论文报告 54.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_gsmplus
论文报告 46.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_bbh
论文报告 31.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_drop
论文报告 35.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_math
论文报告 14.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_tqa
论文报告 54.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_nq
论文报告 23.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_sqa
论文报告 8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_mmlu
论文报告 36.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_mmlu_p
论文报告 18.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_arc_c
论文报告 55 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_obqa
论文报告 56.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_wino
论文报告 51.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_agi
论文报告 35.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_13b_fkd_ife
论文报告 64.9 percentage_points
实际观测 — percentage_points
Assessment（0）
这条 Claim 暂无已发布的不可变 Assessment。
关联运行（0）
这条 Claim 暂无关联执行。