Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for 13B FKD post-training performance.
Source: source_paper:Table 2, Page 9
Reported and observed measurements
downstream_accuracy
t2_13b_fkd_gsm8k
Reported 72.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_gsm_s
Reported 54.3 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_gsmplus
Reported 46.5 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_bbh
Reported 31.4 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_drop
Reported 35.7 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_math
Reported 14.6 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_tqa
Reported 54.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_nq
Reported 23.2 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_sqa
Reported 8 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_mmlu
Reported 36.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_mmlu_p
Reported 18.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_arc_c
Reported 55 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_obqa
Reported 56.4 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_wino
Reported 51.2 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_agi
Reported 35.9 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_fkd_ife
Reported 64.9 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.