{"id":"W4389520096","doi":"10.18653/v1/2023.findings-emnlp.500","title":"Co-training and Co-distillation for Quality Improvement and Compression of Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Computer science; Margin (machine learning); Benchmark (surveying); Distillation; Inference; Language model; Performance improvement; Machine learning; Artificial intelligence; Performance prediction; Simulation; Engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.0003486664,0.00004385424,0.00009123308,0.00003636922,0.00004526016,0.00002682344,0.00007516986,0.0000218778,0.000001408448],"category_scores_gemma":[0.00001466989,0.00003718842,0.0000115158,0.00004319733,0.00001454666,0.0001437989,0.00005469042,0.00002025025,2.894505e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000005340479,"about_ca_system_score_gemma":0.000009690686,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006731112,"about_ca_topic_score_gemma":0.000006003383,"domain_scores_codex":[0.9994818,0.00001441398,0.0001473971,0.0001665915,0.0001000615,0.00008973308],"domain_scores_gemma":[0.9996552,0.0001048073,0.00004916353,0.0001397011,0.00002132248,0.00002974959],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001432426,0.00002076273,0.001182349,0.0002797759,0.00001770701,0.000001027,0.02535398,0.00660446,0.1284336,0.1834333,0.0002869283,0.6543717],"study_design_scores_gemma":[0.0002696134,0.00002892466,0.001091897,0.00001225249,0.000001185717,3.331396e-7,0.0005787027,0.9856645,0.005824375,0.006438763,0.00003851194,0.00005093571],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4076937,0.00001931085,0.591469,0.00008433162,0.00001944239,0.0001021483,0.000004351395,0.0000510323,0.0005566759],"genre_scores_gemma":[0.9689558,0.000005172054,0.03085784,0.00004565949,0.00001241779,0.000006226514,0.000005575709,0.000002412991,0.0001089249],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9790601,"threshold_uncertainty_score":0.15165,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1113164359390328,"score_gpt":0.3746026075313697,"score_spread":0.2632861715923369,"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."}}