Downstream task evaluation after post-training stage DPO for 13b_rkd.
来源:source_paper:Table 8, Page 33, After DPO
报告指标与观测值
downstream_accuracy
t8_dpo_13b_rkd_gsm8k
论文报告 65.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_gsm_s
论文报告 41.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_gsmplus
论文报告 35 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_bbh
论文报告 33.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_drop
论文报告 39.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_math
论文报告 7.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_tqa
论文报告 53.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_nq
论文报告 23.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_sqa
论文报告 7.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_mmlu
论文报告 47.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_mmlu_p
论文报告 17.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_arc_c
论文报告 57.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_obqa
论文报告 56.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_wino
论文报告 51.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_agi
论文报告 38.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_ife
论文报告 64.5 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。