Downstream task evaluation after post-training stage RLVR1 for 7b_rkd.
来源：source_paper:Table 8, Page 33, After RLVR1
报告指标与观测值
downstream_accuracy
t8_rlvr1_7b_rkd_gsm8k
论文报告 77.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_gsm_s
论文报告 54.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_gsmplus
论文报告 52.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_bbh
论文报告 34.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_drop
论文报告 40.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_math
论文报告 15.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_tqa
论文报告 52.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_nq
论文报告 21.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_sqa
论文报告 7.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_mmlu
论文报告 45.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_mmlu_p
论文报告 16.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_arc_c
论文报告 57.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_obqa
论文报告 50.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_wino
论文报告 51.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_agi
论文报告 36.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_7b_rkd_ife
论文报告 59.3 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。