Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for NTP post-training baseline performance.
来源：source_paper:Table 2, Page 9
报告指标与观测值
downstream_accuracy
t2_ntp_shared_gsm8k
论文报告 67 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_gsm_s
论文报告 46.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_gsmplus
论文报告 40.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_bbh
论文报告 31.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_drop
论文报告 32.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_math
论文报告 12.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_tqa
论文报告 53.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_nq
论文报告 21.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_sqa
论文报告 7.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_mmlu
论文报告 32.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_mmlu_p
论文报告 16.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_arc_c
论文报告 52.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_obqa
论文报告 52.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_wino
论文报告 51.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_agi
论文报告 32.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_ntp_shared_ife
论文报告 62.1 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。