Switch Distillation (SD, q=20%, tau=2) downstream evaluation performance after mid-training with OLMo-2 13B Instruct teacher.
来源：source_paper:Table 1, Page 9
报告指标与观测值
downstream_accuracy
t1_13b_sd_gsm8k
论文报告 66 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_gsm_s
论文报告 54.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_gsmplus
论文报告 42.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_bbh
论文报告 31.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_drop
论文报告 45.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_math
论文报告 12 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_tqa
论文报告 54.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_nq
论文报告 24.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_sqa
论文报告 9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_mmlu
论文报告 48.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_mmlu_p
论文报告 17.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_arc_c
论文报告 58.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_obqa
论文报告 63.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_wino
论文报告 51.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t1_13b_sd_agi
论文报告 39.2 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。