Forward KL Distillation (FKD, alpha=0.5) downstream evaluation performance after mid-training with OLMo-2 13B Instruct teacher.
Source: source_paper:Table 1, Page 9
Reported and observed measurements
downstream_accuracy
t1_13b_fkd_gsm8k
Reported 52.7 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_gsm_s
Reported 37.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_gsmplus
Reported 31.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_bbh
Reported 29.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_drop
Reported 33.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_math
Reported 6.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_tqa
Reported 54.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_nq
Reported 24.7 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_sqa
Reported 8.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_mmlu
Reported 47.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_mmlu_p
Reported 16.3 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_arc_c
Reported 55.3 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_obqa
Reported 56.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_wino
Reported 51.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_fkd_agi
Reported 36.2 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.