Full downstream results after standard 4-stage post-training pipeline (SFT, DPO, RLVR1, RLVR2) for 13B RKD post-training performance.
Source: source_paper:Table 2, Page 9
Reported and observed measurements
downstream_accuracy
t2_13b_rkd_gsm8k
Reported 75.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_gsm_s
Reported 58.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_gsmplus
Reported 49.3 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_bbh
Reported 33.3 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_drop
Reported 38.7 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_math
Reported 16 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_tqa
Reported 53.9 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_nq
Reported 22.3 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_sqa
Reported 7.7 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_mmlu
Reported 38.2 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_mmlu_p
Reported 17 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_arc_c
Reported 56 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_obqa
Reported 55.2 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_wino
Reported 51.1 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_agi
Reported 36.6 percentage_points
Observed — percentage_points
downstream_accuracy
t2_13b_rkd_ife
Reported 64.1 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.