Downstream task evaluation after post-training stage DPO for 13b_trkd.
来源:source_paper:Table 8, Page 33, After DPO
报告指标与观测值
downstream_accuracy
t8_dpo_13b_trkd_gsm8k
论文报告 53.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_gsm_s
论文报告 34.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_gsmplus
论文报告 30.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_bbh
论文报告 32.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_drop
论文报告 35.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_math
论文报告 6.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_tqa
论文报告 54.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_nq
论文报告 22.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_sqa
论文报告 8.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_mmlu
论文报告 43.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_mmlu_p
论文报告 16.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_arc_c
论文报告 53.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_obqa
论文报告 56 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_wino
论文报告 52 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_agi
论文报告 35.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_dpo_13b_trkd_ife
论文报告 61.7 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。