Downstream task evaluation after post-training stage DPO for 13b_rkd.
Source: source_paper:Table 8, Page 33, After DPO
Reported and observed measurements
downstream_accuracy
t8_dpo_13b_rkd_gsm8k
Reported 65.6 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_gsm_s
Reported 41.2 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_gsmplus
Reported 35 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_bbh
Reported 33.2 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_drop
Reported 39.9 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_math
Reported 7.4 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_tqa
Reported 53.5 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_nq
Reported 23.7 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_sqa
Reported 7.9 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_mmlu
Reported 47.8 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_mmlu_p
Reported 17.9 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_arc_c
Reported 57.1 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_obqa
Reported 56.2 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_wino
Reported 51.9 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_agi
Reported 38.5 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_13b_rkd_ife
Reported 64.5 percentage_points
Observed — percentage_points
Assessments (0)
No immutable Assessment has been published for this Claim yet.
Runs (0)
No execution has been linked to this Claim yet.