Downstream task evaluation after post-training stage DPO for ntp_shared.
Source: source_paper:Table 8, Page 33, After DPO
Reported and observed measurements
downstream_accuracy
t8_dpo_ntp_shared_gsm8k
Reported 52.4 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_gsm_s
Reported 33.4 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_gsmplus
Reported 28.2 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_bbh
Reported 32.3 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_drop
Reported 33.5 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_math
Reported 6.6 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_tqa
Reported 55.4 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_nq
Reported 21.5 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_sqa
Reported 7.7 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_mmlu
Reported 41.1 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_mmlu_p
Reported 16.4 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_arc_c
Reported 50.3 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_obqa
Reported 50.6 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_wino
Reported 51.5 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_agi
Reported 34.1 percentage_points
Observed — percentage_points
downstream_accuracy
t8_dpo_ntp_shared_ife
Reported 59.9 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.