Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for 7B Switch Distillation post-training performance.
来源:source_paper:Table 2, Page 9
报告指标与观测值
downstream_accuracy
t2_7b_sd_gsm8k
论文报告 79.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_gsm_s
论文报告 65 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_gsmplus
论文报告 52.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_bbh
论文报告 35.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_drop
论文报告 48.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_math
论文报告 22.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_tqa
论文报告 53.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_nq
论文报告 22.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_sqa
论文报告 8.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_mmlu
论文报告 48.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_mmlu_p
论文报告 19.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_arc_c
论文报告 62.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_obqa
论文报告 59.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_wino
论文报告 54.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_agi
论文报告 39.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t2_7b_sd_ife
论文报告 69.5 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。