{"id":"W4391157657","doi":"10.48550/arxiv.2401.11323","title":"Identifying and Analyzing Performance-Critical Tokens in Large Language Models","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Natural Science Foundation of China; Canadian Institute for Advanced Research","keywords":"Computer science; Task (project management); Encoding (memory); Security token; Context (archaeology); Natural language processing; Artificial intelligence","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":[0.003705596,0.0009712569,0.0008427817,0.0008139597,0.0004473779,0.001332887,0.001185475,0.00105572,0.001716856],"category_scores_gemma":[0.03358674,0.0004568706,0.0005251161,0.0006430148,0.0009965027,0.003476783,0.002415779,0.00241618,0.0006655593],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007924205,"about_ca_system_score_gemma":0.001043208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001792003,"about_ca_topic_score_gemma":0.003071344,"domain_scores_codex":[0.997822,0.001085456,0.00009749946,0.0004910528,0.0003073701,0.0001965095],"domain_scores_gemma":[0.977375,0.01764473,0.001488748,0.002123304,0.000761908,0.0006062885],"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":[0.001523218,0.0006278757,0.05530625,0.00135079,0.0002651041,0.0009048363,0.002120579,0.3858501,0.09728026,0.02953366,0.007576752,0.4176606],"study_design_scores_gemma":[0.00002320487,0.0002429987,0.005131822,0.00003402182,0.00003099977,0.0001115759,0.0002286385,0.9457163,0.01658919,0.03040502,0.001449727,0.00003651788],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4973822,0.0007004181,0.4948432,0.0007066933,0.00009304256,0.0000982018,0.0005899423,0.00344377,0.002142516],"genre_scores_gemma":[0.9470614,0.0001254834,0.0509235,0.00008128937,0.00003001509,0.00008070732,0.0007802112,0.0002549019,0.0006624542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003705596,"threshold_uncertainty_score":0.01959729,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08114905561307885,"score_gpt":0.2254104830787246,"score_spread":0.1442614274656457,"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."}}