{"id":"W4396661957","doi":"10.1051/epjconf/202429507024","title":"HEPScore: A new CPU benchmark for the WLCG","year":2024,"lang":"en","type":"article","venue":"EPJ Web of Conferences","topic":"Parallel Computing and Optimization Techniques","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute of Particle Physics; University of Victoria","funders":"","keywords":"Benchmark (surveying); National laboratory; Resource (disambiguation); Software; Procurement; Computer science; Operating system; Physics; Business","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":[0.0003058467,0.00009283896,0.000125062,0.00007793314,0.00008138041,0.0002550574,0.0008860245,0.00003992984,0.00005130069],"category_scores_gemma":[0.00005927471,0.00005909951,0.00008104175,0.0002805428,0.00005843081,0.0001557339,0.0001055707,0.00007511881,0.000007373837],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00000627413,"about_ca_system_score_gemma":0.0009647495,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006637337,"about_ca_topic_score_gemma":0.0000115507,"domain_scores_codex":[0.9992505,0.0000263935,0.000192452,0.0002211838,0.0001651918,0.0001442465],"domain_scores_gemma":[0.9990243,0.0005131157,0.00005604986,0.0002835556,0.00008181583,0.000041174],"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":[0.000004289137,0.00001651168,0.0001323872,0.00003282765,0.00003598867,0.000001223107,0.0005275733,0.0004686299,0.00006877969,0.7743286,0.07901882,0.1453644],"study_design_scores_gemma":[0.00009629593,0.0001464439,0.0002118863,0.0001014299,0.00001037719,0.000003669001,0.00002614498,0.7176254,0.002153483,0.04542919,0.2340764,0.0001192967],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0003194771,0.002252101,0.9792196,0.004374291,0.0005242469,0.0001790444,0.000002580744,0.0003227454,0.0128059],"genre_scores_gemma":[0.9155226,0.0002097206,0.08265328,0.0001438657,0.0001189886,0.00001720402,0.00000201238,0.00000426884,0.001328026],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9152032,"threshold_uncertainty_score":0.2459525,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03219257986751806,"score_gpt":0.29030002148987,"score_spread":0.2581074416223519,"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."}}