Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall
Not assessedPlan blockedMeasurementclaim_t8_rlvr1_7b_fkd

Downstream task evaluation after post-training stage RLVR1 for 7b_fkd.

Source: source_paper:Table 8, Page 33, After RLVR1

Reported and observed measurements

downstream_accuracy

t8_rlvr1_7b_fkd_gsm8k

Reported 76.7 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_gsm_s

Reported 59.4 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_gsmplus

Reported 51.9 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_bbh

Reported 34.4 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_drop

Reported 39.2 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_math

Reported 16.8 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_tqa

Reported 52.8 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_nq

Reported 23.1 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_sqa

Reported 8.3 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_mmlu

Reported 41.4 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_mmlu_p

Reported 18.3 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_arc_c

Reported 59.4 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_obqa

Reported 56.2 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_wino

Reported 51.2 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_agi

Reported 38 percentage_points

Observed — percentage_points

downstream_accuracy

t8_rlvr1_7b_fkd_ife

Reported 67.1 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.