{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004037143,0.00009432813,0.0001175564,0.00009696854,0.0001581326,0.0002210479,0.0003262484,0.00003457582,0.000003784128],"category_scores_gemma":[0.0001081996,0.00008832929,0.00004799293,0.0002096488,0.0000112033,0.0009375255,0.0001994455,0.00007463327,0.000009486753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005148412,"about_ca_system_score_gemma":0.00002093206,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000837773,"about_ca_topic_score_gemma":0.000009156077,"domain_scores_codex":[0.9989868,0.00004027413,0.0001864576,0.0003839918,0.0001528641,0.0002496238],"domain_scores_gemma":[0.9992802,0.0001636595,0.00007550904,0.0003234308,0.0001014682,0.0000557341],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001014472,0.00001799411,0.0004791754,0.00007927538,0.00002426876,0.00001516676,0.04075399,0.1221418,0.005983296,0.6026295,0.0001161395,0.2277492],"study_design_scores_gemma":[0.0001611309,0.00001161218,0.0003937614,0.00004761806,0.000002883409,0.000001630964,0.002671456,0.9822776,0.0003954316,0.01389491,0.00002690174,0.0001150725],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1318844,0.00004372149,0.8638052,0.0003368104,0.0001763123,0.000141164,0.000002338698,0.0005691984,0.003040837],"genre_scores_gemma":[0.9147458,0.000002088783,0.08461264,0.0001227919,0.000299985,0.00001147188,0.000006812208,0.00001164403,0.0001867888],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8601358,"threshold_uncertainty_score":0.3601964,"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."}}