Switch Distillation (SD, q=20%, tau=2) 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_sd_gsm8k
Reported 66 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_gsm_s
Reported 54.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_gsmplus
Reported 42.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_bbh
Reported 31.6 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_drop
Reported 45.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_math
Reported 12 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_tqa
Reported 54.1 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_nq
Reported 24.9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_sqa
Reported 9 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_mmlu
Reported 48.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_mmlu_p
Reported 17.7 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_arc_c
Reported 58.5 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_obqa
Reported 63.8 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_wino
Reported 51.2 percentage_points
Observed — percentage_points
downstream_accuracy
t1_13b_sd_agi
Reported 39.2 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.