Downstream task performance and group macro-averages for Switch Distillation under routing threshold sweep 13b_q10.
Source: source_paper:Table 4, Page 21
Reported and observed measurements
downstream_accuracy
t4_13b_q10_gsm8k
Reported 62.1 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_gsm_s
Reported 49.6 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_gsmplus
Reported 38.3 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_bbh
Reported 28.6 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_drop
Reported 44.2 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_math
Reported 7.8 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_avg_reasoning
Reported 38.5 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_tqa
Reported 52.2 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_nq
Reported 23.1 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_sqa
Reported 9.2 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_avg_factual
Reported 28.1 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_mmlu
Reported 46.8 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_mmlu_p
Reported 17.1 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_arc_c
Reported 55.5 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_obqa
Reported 59.6 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_wino
Reported 52.7 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_agi
Reported 37.9 percentage_points
Observed — percentage_points
downstream_accuracy
t4_13b_q10_avg_knowledge
Reported 45 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.