{"id":"W4416037156","doi":"10.18653/v1/2025.emnlp-industry.134","title":"Deploying Tiny LVLM Judges for Real-World Evaluation of Chart Models: Lessons Learned and Best Practices","year":2025,"lang":"","type":"article","venue":"","topic":"Machine Learning and Data Classification","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Compute Canada","keywords":"Best practice; Chart; Natural language; Natural (archaeology); Column (typography)","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":[0.008216922,0.002806753,0.001778218,0.001986513,0.0009780304,0.003424136,0.005625587,0.002456335,0.01206248],"category_scores_gemma":[0.02680107,0.0009486178,0.001021484,0.001193086,0.0007873403,0.0056874,0.002870245,0.002868601,0.008358074],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001884146,"about_ca_system_score_gemma":0.001386213,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01220818,"about_ca_topic_score_gemma":0.02171594,"domain_scores_codex":[0.9946226,0.002144494,0.0004787807,0.001161115,0.001268049,0.0003249813],"domain_scores_gemma":[0.9887766,0.005307703,0.0002896602,0.002131962,0.0029498,0.0005442661],"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":[0.001847055,0.001179432,0.005648868,0.001022965,0.0007188071,0.0006864498,0.0004742598,0.1197298,0.02549071,0.002627764,0.1164412,0.7241327],"study_design_scores_gemma":[0.0002568783,0.0002736631,0.001025111,0.0001313286,0.00007199238,0.0001129614,0.0003094162,0.961684,0.0193534,0.004758214,0.01194138,0.00008156024],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1373949,0.007098888,0.5682134,0.002810999,0.003542647,0.001114911,0.006854954,0.2586154,0.01435388],"genre_scores_gemma":[0.518218,0.001239306,0.4417903,0.001448143,0.0003961217,0.0008645313,0.01834489,0.01054565,0.007153197],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01220818,"threshold_uncertainty_score":0.04345572,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3862536639501394,"score_gpt":0.4808505296975183,"score_spread":0.09459686574737886,"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."}}