{"id":"W4389519204","doi":"10.18653/v1/2023.emnlp-main.721","title":"Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"","keywords":"Computer science; Language model; Decoding methods; Overhead (engineering); Cache; Artificial intelligence; Text generation; Algorithm; Parallel computing; Programming language","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":[0.001574857,0.001009192,0.0008134197,0.0005110598,0.0004121812,0.000934768,0.001484179,0.001182829,0.004213699],"category_scores_gemma":[0.01157309,0.0004036722,0.0005386141,0.0004862955,0.0006875087,0.001581228,0.001394578,0.001530883,0.002658313],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005809521,"about_ca_system_score_gemma":0.001200944,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002230017,"about_ca_topic_score_gemma":0.003746161,"domain_scores_codex":[0.9988831,0.0004938067,0.00006935284,0.0002444372,0.000227942,0.0000815126],"domain_scores_gemma":[0.9953415,0.003314702,0.0001893378,0.0005454158,0.0004871282,0.0001218969],"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":[0.0005844249,0.000243229,0.002356676,0.0002983322,0.00006623908,0.0004312753,0.0005506726,0.4209057,0.02929158,0.02554112,0.01341763,0.5063131],"study_design_scores_gemma":[0.00005022593,0.0000478205,0.00008832715,0.00000948552,0.00001115089,0.00005122779,0.00001616813,0.9820961,0.006420872,0.009274275,0.001920198,0.00001421083],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02577787,0.0002621289,0.9609732,0.0003580333,0.0001287326,0.000167216,0.0002454434,0.008841256,0.003246017],"genre_scores_gemma":[0.5207698,0.0001931935,0.4676034,0.0004233047,0.0001202811,0.0004641756,0.0009460065,0.001797627,0.007682318],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004213699,"threshold_uncertainty_score":0.0140962,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04398754862890349,"score_gpt":0.256004301114469,"score_spread":0.2120167524855655,"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."}}