Downstream task evaluation after post-training stage RLVR1 for 13b_sd.
来源：source_paper:Table 8, Page 33, After RLVR1
报告指标与观测值
downstream_accuracy
t8_rlvr1_13b_sd_gsm8k
论文报告 78.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_gsm_s
论文报告 65.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_gsmplus
论文报告 52.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_bbh
论文报告 33.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_drop
论文报告 43.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_math
论文报告 17.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_tqa
论文报告 54.3 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_nq
论文报告 24 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_sqa
论文报告 8.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_mmlu
论文报告 38.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_mmlu_p
论文报告 17.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_arc_c
论文报告 56.2 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_obqa
论文报告 57.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_wino
论文报告 51.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_agi
论文报告 37.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_rlvr1_13b_sd_ife
论文报告 68.9 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。