Downstream task evaluation after post-training stage SFT for ntp_shared.
来源:source_paper:Table 8, Page 33, After SFT
报告指标与观测值
downstream_accuracy
t8_sft_ntp_shared_gsm8k
论文报告 45.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_gsm_s
论文报告 28.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_gsmplus
论文报告 23.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_bbh
论文报告 30.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_drop
论文报告 33.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_math
论文报告 6.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_tqa
论文报告 56.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_nq
论文报告 21.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_sqa
论文报告 8.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_mmlu
论文报告 39.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_mmlu_p
论文报告 16.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_arc_c
论文报告 51.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_obqa
论文报告 51.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_wino
论文报告 51.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_agi
论文报告 33.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_ntp_shared_ife
论文报告 45.8 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。