Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall
尚未评估计划受阻指标结果claim_t8_dpo_7b_sd

Downstream task evaluation after post-training stage DPO for 7b_sd.

来源:source_paper:Table 8, Page 33, After DPO

报告指标与观测值

downstream_accuracy

t8_dpo_7b_sd_gsm8k

论文报告 70.9 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_gsm_s

论文报告 50.8 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_gsmplus

论文报告 44.2 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_bbh

论文报告 35.5 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_drop

论文报告 49 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_math

论文报告 15 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_tqa

论文报告 54.7 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_nq

论文报告 23.6 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_sqa

论文报告 8.1 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_mmlu

论文报告 50 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_mmlu_p

论文报告 20.1 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_arc_c

论文报告 61.2 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_obqa

论文报告 61 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_wino

论文报告 55.5 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_agi

论文报告 40.9 percentage_points

实际观测 — percentage_points

downstream_accuracy

t8_dpo_7b_sd_ife

论文报告 62.1 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。