Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for 7B RKD post-training performance.
来源：source_paper:Table 2, Page 9
报告指标与观测值
downstream_accuracy
t2_7b_rkd_gsm8k
论文报告 73.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_gsm_s
论文报告 53.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_gsmplus
论文报告 49.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_bbh
论文报告 31.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_drop
论文报告 40.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_math
论文报告 17.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_tqa
论文报告 51.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_nq
论文报告 21.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_sqa
论文报告 7.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_mmlu
论文报告 46.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_mmlu_p
论文报告 18 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_arc_c
论文报告 61.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_obqa
论文报告 53 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_wino
论文报告 51.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_agi
论文报告 37.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_rkd_ife
论文报告 61.6 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。