Downstream task evaluation after post-training stage SFT for 13b_rkd.
来源:source_paper:Table 8, Page 33, After SFT
报告指标与观测值
downstream_accuracy
t8_sft_13b_rkd_gsm8k
论文报告 56.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_gsm_s
论文报告 36.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_gsmplus
论文报告 31.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_bbh
论文报告 32.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_drop
论文报告 38.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_math
论文报告 11.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_tqa
论文报告 55.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_nq
论文报告 23.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_sqa
论文报告 8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_mmlu
论文报告 47.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_mmlu_p
论文报告 17.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_arc_c
论文报告 57.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_obqa
论文报告 57.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_wino
论文报告 52.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_agi
论文报告 38.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_rkd_ife
论文报告 43.6 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。