Downstream task evaluation after post-training stage SFT for 7b_trkd.
来源：source_paper:Table 8, Page 33, After SFT
报告指标与观测值
downstream_accuracy
t8_sft_7b_trkd_gsm8k
论文报告 49.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_gsm_s
论文报告 31.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_gsmplus
论文报告 26.9 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_bbh
论文报告 31.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_drop
论文报告 37.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_math
论文报告 7.6 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_tqa
论文报告 53.7 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_nq
论文报告 23.1 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_sqa
论文报告 8.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_mmlu
论文报告 46.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_mmlu_p
论文报告 17.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_arc_c
论文报告 57.8 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_obqa
论文报告 56 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_wino
论文报告 51.4 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_agi
论文报告 37.5 percentage_points
实际观测 — percentage_points
downstream_accuracy
t8_sft_7b_trkd_ife
论文报告 45.3 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。