Downstream task evaluation after post-training stage SFT for 13b_fkd.
来源:source_paper:Table 8, Page 33, After SFT
报告指标与观测值
downstream_accuracy
t8_sft_13b_fkd_gsm8k
论文报告 54.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_gsm_s
论文报告 32.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_gsmplus
论文报告 28 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_bbh
论文报告 31.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_drop
论文报告 36.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_math
论文报告 7.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_tqa
论文报告 54.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_nq
论文报告 23.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_sqa
论文报告 7.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_mmlu
论文报告 45.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_mmlu_p
论文报告 17.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_arc_c
论文报告 55.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_obqa
论文报告 57.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_wino
论文报告 51.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_agi
论文报告 37.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_13b_fkd_ife
论文报告 44.2 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。