Reverse KL Distillation (RKD, 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_rkd_gsm8k
Reported 62.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_gsm_s
Reported 46.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_gsmplus
Reported 39.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_bbh
Reported 31.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_drop
Reported 43.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_math
Reported 10.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_tqa
Reported 53.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_nq
Reported 24.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_sqa
Reported 8.4 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_mmlu
Reported 50.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_mmlu_p
Reported 18.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_arc_c
Reported 62.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_obqa
Reported 62.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_wino
Reported 52.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_7b_rkd_agi
Reported 39.3 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.