{"id":"W4386566583","doi":"10.18653/v1/2023.findings-eacl.12","title":"Revisiting Intermediate Layer Distillation for Compressing Language Models: An Overfitting Perspective","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"Institute for Information and Communications Technology Promotion; Ministry of Science and ICT, South Korea; Korea Advanced Institute of Science and Technology","keywords":"Overfitting; Computer science; Distillation; Code (set theory); Benchmark (surveying); Machine learning; Artificial intelligence; Consistency (knowledge bases); Simple (philosophy); Transformer; Layer (electronics); Language model; Natural language processing; Artificial neural network; Programming language; Engineering; Chemistry","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":[0.002620301,0.001642673,0.001374876,0.00125189,0.0007569592,0.002129805,0.002835578,0.001462249,0.004452685],"category_scores_gemma":[0.01406888,0.0006521263,0.001337629,0.00174444,0.001422903,0.005099816,0.003922566,0.00454647,0.002504646],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008820301,"about_ca_system_score_gemma":0.002722203,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005960368,"about_ca_topic_score_gemma":0.0116382,"domain_scores_codex":[0.9988434,0.0003872972,0.00009233844,0.0002778756,0.0002665518,0.000132502],"domain_scores_gemma":[0.996595,0.001877034,0.000152797,0.0008732528,0.0003798345,0.0001221744],"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":[0.0004113546,0.000227812,0.002616694,0.0004211397,0.0001868658,0.0002631263,0.0003293504,0.315048,0.01330687,0.03222757,0.01150181,0.6234594],"study_design_scores_gemma":[0.00003209621,0.00006261216,0.0002050896,0.00003018676,0.00002511247,0.00007555249,0.00005198607,0.9682076,0.00664079,0.02139561,0.003248967,0.00002445506],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02387727,0.0007869448,0.9678189,0.0006374061,0.0001006305,0.00008730482,0.0004516497,0.004644956,0.001594887],"genre_scores_gemma":[0.4038042,0.0008486578,0.581634,0.001145342,0.0002180176,0.0003447561,0.003831493,0.001302761,0.006870615],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005960368,"threshold_uncertainty_score":0.01489568,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0662106754848642,"score_gpt":0.3361705680851451,"score_spread":0.2699598926002809,"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."}}