{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009689009,0.0011753,0.0004295472,0.0005515697,0.0003153202,0.001220567,0.002076823,0.0006291838,0.008724901],"category_scores_gemma":[0.004933278,0.0008681394,0.001274047,0.0004639361,0.0008965236,0.002618913,0.002379198,0.002419665,0.00392542],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008913318,"about_ca_system_score_gemma":0.001860193,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002522631,"about_ca_topic_score_gemma":0.003431903,"domain_scores_codex":[0.9993393,0.0001006566,0.0000549891,0.0001603407,0.0002294407,0.0001152081],"domain_scores_gemma":[0.9985657,0.0005313682,0.0000885769,0.0004990552,0.0002384014,0.00007691645],"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":[0.002314518,0.0006097047,0.01474207,0.001609058,0.0003383563,0.0009575221,0.001671832,0.1201445,0.06525236,0.1098415,0.212415,0.4701037],"study_design_scores_gemma":[0.0003933182,0.0002435305,0.002747403,0.0001662925,0.0001072742,0.0004175587,0.0001672664,0.6996123,0.1058873,0.05885148,0.1312702,0.0001361743],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.04066892,0.0004820532,0.5968153,0.0003837016,0.0003314409,0.0001583388,0.001392451,0.3457431,0.01402463],"genre_scores_gemma":[0.3254541,0.0005852345,0.5450793,0.0009279784,0.0001253686,0.0005764842,0.006376704,0.1050002,0.01587461],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.008724901,"threshold_uncertainty_score":0.02918768,"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."}}