{"id":"W4385570159","doi":"10.18653/v1/2023.findings-acl.714","title":"A Study on Knowledge Distillation from Weak Teacher for Scaling Up Pre-trained Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Initialization; Computer science; Artificial intelligence; Weighting; Distillation; Machine learning; Language model; Domain (mathematical analysis); Natural language processing; Mathematics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003323753,0.001150812,0.0008606329,0.0006032265,0.0004792449,0.001248088,0.001574577,0.001096195,0.002111859],"category_scores_gemma":[0.01826956,0.0006309454,0.0007337781,0.0008040514,0.0006880295,0.004180647,0.001655289,0.002671397,0.0005911286],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009882589,"about_ca_system_score_gemma":0.001451414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006189129,"about_ca_topic_score_gemma":0.006056724,"domain_scores_codex":[0.99901,0.0003887283,0.00008550315,0.0002364544,0.0001989292,0.00008044573],"domain_scores_gemma":[0.9937884,0.004112816,0.0002027015,0.0008883737,0.0008164687,0.0001911515],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0007926587,0.0006570501,0.005227523,0.0005476104,0.0001702207,0.0002259804,0.0005979614,0.3953724,0.03410161,0.009382524,0.003055354,0.5498691],"study_design_scores_gemma":[0.00005240114,0.0003265852,0.0005873047,0.00002269472,0.00003728381,0.00006611811,0.00006686209,0.9795306,0.01552254,0.001621919,0.002152407,0.00001332424],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3110614,0.002287572,0.6740603,0.001200369,0.0001917711,0.0003467106,0.0001806106,0.004428124,0.006243194],"genre_scores_gemma":[0.6623272,0.0006469581,0.3321487,0.0002960572,0.00007191708,0.000169273,0.0005389426,0.0005063824,0.003294529],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006189129,"threshold_uncertainty_score":0.01757789,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0692850980979679,"score_gpt":0.3390118085480475,"score_spread":0.2697267104500796,"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."}}