{"id":"W6966641450","doi":"10.48448/gfnv-vm46","title":"Universal-KD: Attention-based Output-Grounded Intermediate Layer Knowledge Distillation","year":2021,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Distillation; Layer (electronics); Matching (statistics); Key (lock); Column (typography); Fractionating column","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001062995,0.0006817162,0.0006296841,0.002269285,0.0003673587,0.0004093422,0.001337587,0.0003976612,0.004193158],"category_scores_gemma":[0.0003981999,0.0006743001,0.0002130513,0.003109032,0.00224056,0.000318064,0.0004242292,0.0005135136,0.005200152],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001482086,"about_ca_system_score_gemma":0.002639432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004878282,"about_ca_topic_score_gemma":0.005483235,"domain_scores_codex":[0.9953372,0.0002193282,0.000554004,0.001580858,0.001375649,0.0009329314],"domain_scores_gemma":[0.9968721,0.0001621432,0.0006855888,0.001228401,0.0006361017,0.0004156731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001522329,0.002475071,0.005762717,0.001042931,0.0006645274,0.000334317,0.0009803034,0.001848756,0.02860691,0.02440391,0.8936781,0.04005026],"study_design_scores_gemma":[0.003064639,0.000182727,0.008055622,0.001910388,0.0003923773,0.00001647795,0.0006887965,0.1236395,0.0004641219,0.0004935896,0.8587753,0.002316485],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.002065591,0.001190056,0.02352663,0.0003256813,0.00510265,0.001187288,0.0007033806,0.001689132,0.9642096],"genre_scores_gemma":[0.47736,0.00002269464,0.003760025,0.0001053425,0.0008777205,0.0000223858,0.001613791,0.0009733989,0.5152647],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.4752944,"threshold_uncertainty_score":0.9995708,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03976833101735353,"score_gpt":0.3152315852717479,"score_spread":0.2754632542543943,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}