{"id":"W4416021351","doi":"10.48550/arxiv.2511.02043","title":"Flashlight: PyTorch Compiler Extensions to Accelerate Attention Variants","year":2025,"lang":"","type":"preprint","venue":"ArXiv.org","topic":"Parallel Computing and Optimization Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; University of Texas at Austin; Georgia Institute of Technology; National Science Foundation","keywords":"Compiler; Workflow; Kernel (algebra); Template; Block (permutation group theory); Flexibility (engineering); Computation; Pipeline (software); Compile time","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001336299,0.001128309,0.00125521,0.001036279,0.001213864,0.0009117129,0.004212066,0.0008573133,0.0002187184],"category_scores_gemma":[0.000302952,0.001216063,0.0005568723,0.002158333,0.0001484941,0.000559454,0.007600995,0.001606514,0.001913576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003191885,"about_ca_system_score_gemma":0.0008014457,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001885686,"about_ca_topic_score_gemma":0.00001896122,"domain_scores_codex":[0.9921635,0.0007688564,0.001768944,0.003164526,0.0008513381,0.001282836],"domain_scores_gemma":[0.9932722,0.0002991697,0.0007579608,0.003461921,0.00157917,0.0006295271],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.0003927899,0.003805525,0.1806452,0.001159908,0.001640543,0.0004585873,0.007173021,0.4097796,0.0127617,0.02620351,0.2944576,0.06152195],"study_design_scores_gemma":[0.0009854229,0.0003382693,0.5163247,0.003014429,0.000190907,0.0000302438,0.00003242769,0.4468563,0.003869575,0.001559819,0.02431101,0.002486872],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1343666,0.0002320514,0.8437019,0.005239063,0.006107431,0.001551083,0.0000558303,0.001427949,0.007318148],"genre_scores_gemma":[0.8184768,0.0007837876,0.155378,0.004602135,0.00070806,0.0002095242,0.0001466752,0.00007682422,0.01961819],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6883239,"threshold_uncertainty_score":0.9990289,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07403998737022514,"score_gpt":0.3224161343961718,"score_spread":0.2483761470259466,"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."}}