Downstream task evaluation after post-training stage RLVR1 for ntp_shared.
来源：source_paper:Table 8, Page 33, After RLVR1
报告指标与观测值
downstream_accuracy
t8_rlvr1_ntp_shared_gsm8k
论文报告 69.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_gsm_s
论文报告 45.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_gsmplus
论文报告 40.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_bbh
论文报告 32.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_drop
论文报告 33.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_math
论文报告 11.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_tqa
论文报告 54.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_nq
论文报告 21.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_sqa
论文报告 8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_mmlu
论文报告 31.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_mmlu_p
论文报告 16.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_arc_c
论文报告 52.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_obqa
论文报告 54.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_wino
论文报告 51.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_agi
论文报告 33.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_ntp_shared_ife
论文报告 63.6 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。