Forward KL Distillation (FKD, alpha=0.5) downstream evaluation performance after mid-training with OLMo-2 7B Instruct teacher.
Source: source_paper:Table 1, Page 9
Reported and observed measurements
downstream_accuracy
t1_7b_fkd_gsm8k
Reported 57.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_gsm_s
Reported 42.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_gsmplus
Reported 36.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_bbh
Reported 30.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_drop
Reported 41.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_math
Reported 7.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_tqa
Reported 53.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_nq
Reported 24.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_sqa
Reported 8.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_mmlu
Reported 49.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_mmlu_p
Reported 18.7 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_arc_c
Reported 61.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_obqa
Reported 61.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_wino
Reported 51.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_fkd_agi
Reported 38.6 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.