{"id":"W4416036669","doi":"10.18653/v1/2025.emnlp-main.368","title":"LEO-MINI: An Efficient Multimodal Large Language Model using Conditional Token Reduction and Mixture of Multi-Modal Experts","year":2025,"lang":"","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Government of Canada","keywords":"Reduction (mathematics); Language model; Security token; Data modeling; Natural language","routes":{"ca_aff":true,"ca_fund":true,"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":[0.001174183,0.001219775,0.001036536,0.0006586431,0.000438385,0.00112328,0.002834696,0.001182146,0.004149754],"category_scores_gemma":[0.003274016,0.0007559381,0.001560542,0.0005943688,0.0005078091,0.002648385,0.002135461,0.002634707,0.00218143],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009738197,"about_ca_system_score_gemma":0.001679476,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008075685,"about_ca_topic_score_gemma":0.01656945,"domain_scores_codex":[0.9994645,0.0001398942,0.00002984935,0.0001856058,0.0001128201,0.00006732353],"domain_scores_gemma":[0.9991191,0.000493484,0.00004699167,0.0001215768,0.000157149,0.00006169156],"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":[0.0007145075,0.0003431556,0.001310635,0.0004097499,0.0003363227,0.0003073645,0.0002837775,0.358028,0.02230971,0.01870332,0.02921244,0.568041],"study_design_scores_gemma":[0.00001610132,0.00003056205,0.00008645724,0.000006400181,0.00001728044,0.0000324711,0.00001601928,0.9908982,0.002458923,0.0048207,0.001602805,0.00001411768],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00712445,0.0003244411,0.9829838,0.0002427151,0.00005840326,0.0001035227,0.0005232801,0.007634145,0.001005126],"genre_scores_gemma":[0.2617607,0.0004364958,0.7196755,0.001059428,0.0001399268,0.0007171743,0.004768587,0.001280102,0.01016211],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.008075685,"threshold_uncertainty_score":0.01605737,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02742399061934513,"score_gpt":0.3145264251647307,"score_spread":0.2871024345453855,"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."}}