{"paper":{"title":"DRAMA: Dissecting Attention Redundancy for Accelerating Multimodal Diffusion Large Language Models","arxiv_id":"2611.16345","primary_cat":"cs.LG","submitted_at":"2026-11-12T06:04:08+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Multimodal diffusion large language models (dLLMs) generate text through iterative masked denoising with bidirectional attention, offering a compelling alternative to autoregressive vision-language modeling.","hedge":{"p":0.97,"label":"asserted"},"own_p":0.05,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.3,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.96,"id":"s1","own":false,"role":"background","text":"However, each denoising step requires a full forward pass over long multimodal sequences in which visual tokens dominate the sequence length, and a fixed step budget further compounds this quadratic per-step cost.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.09,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.19,"role_probabilities":{"gap":0.03,"method":0.0,"result":0.0,"background":0.96,"limitation":0.01,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.98,"id":"s2","own":false,"role":"gap","text":"Existing inference acceleration approaches for multimodal dLLMs characterize attention redundancy implicitly, motivating designs from isolated attention visualizations rather than from measurement.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.05,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.98,"method":0.0,"result":0.0,"background":0.02,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"contribution","text":"We close this gap with DRAMA, a training-free acceleration framework grounded in a systematic dissection of attention redundancy in multimodal dLLMs.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.8,"main_claim_p":0.08,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":1.0,"interpretation":0.0}},{"i":4,"p":0.98,"id":"s4","own":true,"role":"result","text":"By probing attention tensors along spatial and temporal axes, we reveal two persistent redundancy structures: (i) attention mass consistently concentrates on a sparse subset of visual tokens, and (ii) attention distributions rapidly stabilize across denoising steps after an initial formation phase.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.16,"quantitative":false,"quantitative_p":0.33,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.98,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.01}},{"i":5,"p":0.67,"id":"s5","own":true,"role":"method","text":"These findings motivate two complementary mechanisms: Spatial-Redundancy-Aware Visual Token Pruning (SVTP) and Sufficiency-Guided Adaptive Stopping (SGAS).","hedge":{"p":0.81,"label":"asserted"},"own_p":0.73,"main_claim_p":0.08,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.67,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.29,"interpretation":0.04}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"Together, DRAMA achieves an average 18.67 times inference speedup across six vision-language benchmarks without training, architectural modification, or attention caching, while maintaining competitive accuracy.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.6,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":0.95,"id":"s7","own":true,"role":"interpretation","text":"Our results establish a strong, measurement-grounded acceleration baseline for multimodal dLLM inference.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.88,"main_claim_p":0.55,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.04,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.95}}]},"compiled":{"main_claim":{"p":0.59,"text":"Our results establish a strong, measurement-grounded acceleration baseline for multimodal dLLM inference.","location":"abstract s7","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":6402,"cost_usd_reported":0.0},"compiled_at":"2026-11-13","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.11,"states_limitation":0.12,"human_participants":0.03,"controlled_comparison":0.36,"randomized_assignment":0.01,"states_quantitative_result":0.99,"compares_with_prior_results":0.46,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":0.98,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.98,"simulation_or_computation":0.02}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.89,"text":"We close this gap with DRAMA, a training-free acceleration framework grounded in a systematic dissection of attention redundancy in multimodal dLLMs.","location":"abstract s3"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":8,"sentences_in_abstract":8},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}