Per-task downstream accuracy for mid-training ablation always_ce using OLMo-2 7B Instruct teacher.
Source: source_paper:Table 9, Page 33
Reported and observed measurements
downstream_accuracy
t9_always_ce_gsm8k
Reported 69.1 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_gsm_s
Reported 55.2 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_gsmplus
Reported 46.5 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_bbh
Reported 33.3 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_drop
Reported 50.1 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_math
Reported 12 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_tqa
Reported 54.3 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_nq
Reported 24.5 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_sqa
Reported 9.4 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_mmlu
Reported 51.1 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_mmlu_p
Reported 19.9 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_arc_c
Reported 63.7 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_obqa
Reported 63.6 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_wino
Reported 52.7 percentage_points
Observed — percentage_points
downstream_accuracy
t9_always_ce_agi
Reported 41.1 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.