Reverse KL Distillation (RKD, 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_rkd_gsm8k
Reported 59.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_gsm_s
Reported 48.7 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_gsmplus
Reported 37.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_bbh
Reported 31.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_drop
Reported 37.3 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_math
Reported 8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_tqa
Reported 53.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_nq
Reported 24 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_sqa
Reported 8.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_mmlu
Reported 48.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_mmlu_p
Reported 17.3 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_arc_c
Reported 57.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_obqa
Reported 60 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_wino
Reported 51 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_rkd_agi
Reported 38.3 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.