{"id":"W4287889735","doi":"10.18653/v1/2022.findings-naacl.151","title":"Great Power, Great Responsibility: Recommendations for Reducing Energy for Training Language Models","year":2022,"lang":"en","type":"article","venue":"Findings of the Association for Computational Linguistics: NAACL 2022","topic":"Topic Modeling","field":"Computer Science","cited_by":60,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Massachusetts Green High Performance Computing Center; Leukemia and Lymphoma Society of Canada; National Science Foundation","keywords":"Computer science; Energy consumption; Inference; Pace; Cloud computing; Transformer; Efficient energy use; Machine learning; Artificial intelligence; Risk analysis (engineering); Engineering","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00952204,0.002044359,0.001177875,0.00290801,0.001869334,0.005841955,0.00773802,0.006517267,0.0493919],"category_scores_gemma":[0.05165201,0.001655877,0.001395437,0.004385248,0.003348167,0.02011098,0.003375087,0.009638011,0.02465106],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002918367,"about_ca_system_score_gemma":0.006886862,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01761981,"about_ca_topic_score_gemma":0.03374566,"domain_scores_codex":[0.9948332,0.002319349,0.000354152,0.0005226181,0.001586221,0.0003844307],"domain_scores_gemma":[0.9603187,0.02444462,0.000822134,0.003493205,0.008396626,0.002524688],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001845282,0.0003196915,0.001132166,0.001031424,0.00006450732,0.0001086145,0.0002975565,0.007133686,0.001585319,0.05173662,0.608946,0.3274601],"study_design_scores_gemma":[0.0003150504,0.0001233464,0.001441344,0.001634711,0.00008641853,0.000231681,0.001020589,0.02840986,0.002972047,0.15743,0.8061858,0.0001492422],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.004426419,0.04596385,0.2954997,0.576948,0.007578787,0.0005660168,0.002152587,0.01364961,0.05321502],"genre_scores_gemma":[0.08480859,0.06035386,0.6762857,0.1006701,0.004559816,0.002252163,0.004844624,0.007806075,0.05841904],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.0493919,"threshold_uncertainty_score":0.1652324,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03769841022498763,"score_gpt":0.2914918612409227,"score_spread":0.2537934510159351,"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."}}