{"id":"fc3b3fa4-1f18-4bae-abff-7924c393be85","arxiv_id":"2611.16345","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A training-free framework prunes visual tokens to one salient patch and stops denoising early, achieving 18.67x faster multimodal dLLM inference at -3.4 pp average accuracy.","lead":"DRAMA speeds up multimodal image-and-text diffusion language models by keeping only the image patches the model actually attends to and stopping denoising early once the answer stops changing, reaching about 18.7 times faster inference on standard benchmarks. It claims this works with no retraining and only a few points of average accuracy loss, though the accuracy cost grows much larger on text-heavy image tasks.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"SGAS stop predicate relies on lexical coherence plus task-schema parseability as a sufficiency surrogate for output correctness; the paper's own TextVQA result shows this assumption breaks on text-dense inputs, and the derivation in Appendix D does not bridge the gap between attention…","rationale":"The reader's weakest_assumption correctly identifies the SGAS sufficiency surrogate as load-bearing. The paper's own TextVQA result (Table 10) provides direct evidence that this assumption breaks on text-dense inputs, with accuracy falling from 64.60 to 52.50 under FG Uni. The Appendix D derivation bounds output drift between consecutive steps but does not connect stationarity to correctness. The speedup measurements are robust and repeatable, but the quality-preservation claim is conditional on the SGAS predicate being a reliable proxy for correctness. The authors acknowledge the limitation (b312), but the TextVQA result shows it is not a minor edge case—it is a systematic failure on a standard benchmark. The MMBench FG Uni outlier (69.0% vs. 81.9% baseline, a 12.9 pp drop) further suggests the surrogate can fail on non-text-dense benchmarks as well. The verdict should remain CONDITIONAL because the speedup claim is solid, but the accuracy claim needs tighter scoping. I agree with the reader's assessment.","tokens_in":31561,"tokens_out":753,"duration_ms":7475,"concrete_test":"Run DRAMA FG Uni on the full TextVQA validation set (or a held-out OCR-heavy benchmark like OCR-VQA) and report accuracy alongside the 64-step baseline. If the accuracy gap exceeds 10 pp (as the 100-sample subset suggests), the SGAS sufficiency surrogate is not reliable for text-dense inputs, and the 'competitive accuracy' claim must be scoped to benchmarks without dense text. Additionally, compute the correlation between SGAS stopping step and per-sample accuracy: if early-stopped samples are disproportionately wrong, the predicate is selecting for fluency, not correctness.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quality-preservation claim of DRAMA depends on the SGAS stop predicate (Eq. 5, Section 3.3.2) treating decoded-text lexical coherence and task-schema parseability as a sufficient surrogate for output correctness. The paper's own held-out TextVQA evaluation (Table 10, b312) shows this assumption fails: accuracy drops from 64.60 to 52.50 under FG Uni, a 12.1 pp loss that the authors acknowledge (b312: 'DRAMA adds no mechanism to recover quality lost to a reduced budget, and on open-ended text-dense inputs that cost is simply larger'). This is not a corner case—TextVQA is a standard benchmark for text-rich visual reasoning. The Appendix D derivation (Eqs. 6–13) shows that when inter-step JSD collapses, layer outputs stabilize, but it does not prove that a stable output is a correct output. The bound on output drift (Eq. 13) bounds the difference between consecutive-step outputs, not the distance to the ground-truth answer. A fluent, parseable but wrong answer satisfies the SGAS predicate and passes the gate. The paper's own data confirms this failure mode on a held-out benchmark. The speedup numbers are measured and repeatable, but the 'competitive accuracy' claim is conditioned on the SGAS surrogate holding, and the authors' own evidence shows it does not hold on text-dense inputs.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper presents DRAMA, a training-free inference acceleration framework for multimodal diffusion large language models (dLLMs). It identifies two forms of attention redundancy: spatial (attention concentrates on a small subset of visual tokens) and temporal (attention distributions stabilize early in the denoising process). Based on these findings, it proposes Spatial-Redundancy-Aware Visual Token Pruning (SVTP) to prune visual tokens before denoising, and Sufficiency-Guided Adaptive Stopping (SGAS) to terminate denoising early when the output is deemed sufficient via lexical coherence and task-schema parseability. The framework achieves an average 18.67x speedup on LLaDA-V across six benchmarks with competitive accuracy, without training or architectural changes.","tokens_in":31989,"tokens_out":1978,"duration_ms":14926,"significance":"If the claims hold, DRAMA provides a substantial practical speedup for an emerging class of models (multimodal dLLMs) that are currently computationally expensive. The systematic measurement of redundancy is a methodological contribution. The speedup numbers are large and measured, not asymptotic. The paper also provides code and detailed ablations. However, the significance is tempered by the fact that the quality-preservation claim depends on a surrogate (lexical coherence + parseability) that the authors' own held-out evaluation shows can fail significantly (TextVQA). The framework is also evaluated on a single backbone (LLaDA-V) with limited generalization to LaViDa (much smaller speedup). The core insight about spatial and temporal redundancy is valuable, but the practical impact depends on how often the SGAS surrogate holds.","major_comments":[{"comment":"The SGAS stop predicate (Eq. 5) uses lexical coherence and task-schema parseability as a sufficiency surrogate for output correctness. The paper's own held-out evaluation on TextVQA (Table 10) shows this assumption fails dramatically: accuracy drops from 64.60 to 52.50 under FG Uni, a 12.1 pp loss. The authors acknowledge this (b312: 'DRAMA adds no mechanism to recover quality lost to a reduced budget, and on open-ended text-dense inputs that cost is simply larger'). This is not a corner case; TextVQA is a standard benchmark for text-rich visual reasoning. The Appendix D derivation (Eqs. 6-13) bounds output drift between consecutive steps, not distance to ground truth. A fluent, parseable but wrong answer passes the gate. The central claim of 'competitive accuracy' is therefore conditioned on the surrogate holding, and the authors' own evidence shows it does not hold on a held-out benchmark. This is a load-bearing issue for the quality-preservation claim.","section":"Section 3.3.2, Table 10","authors_acknowledge":"As stated in Appendix A.1, DRAMA adds no mechanism to recover quality lost to a reduced budget, and on open-ended text-dense inputs that cost is simply larger than on the six main benchmarks. 19 D Derivation of the Output-Stationarity Bound This appendix supplies the derivation of the Pinsker-type output-stationarity bound invoked in Section 3.3.2, showing that the per-step layer-output drift vanishes as a direct corollary of the JSD collapse identified in Section 3.2.","authors_acknowledge_p":0.72},{"comment":"All accuracy results in Table 1 are reported as single-run point estimates with no error bars, confidence intervals, or significance tests. Given the variability inherent in LLM evaluation, it is impossible to assess whether the reported differences (e.g., -1.9 pp, -3.4 pp) are statistically significant or within noise. This is particularly concerning for claims like 'DRAMA variants substantially outperform Baseline 16' and 'on MME, every DRAMA variant surpasses the unaccelerated baseline (78.9% vs. 78.5%)'. The paper should report standard errors or confidence intervals, or at minimum state whether multiple runs were performed.","section":"Table 1, Table 10","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"SVTP selects patches based on cosine similarity between mean-pooled patch embeddings and mean-pooled question embeddings. This is a heuristic that is not validated against the actual attention distributions that SVTP is meant to approximate. The paper shows that attention concentrates on a few patches, but does not show that the cosine similarity score is a good proxy for attention salience. A direct comparison (e.g., overlap between top-k patches by similarity vs. top-k by attention) is missing. Without this, the claim that SVTP 'grounds' pruning in the measured redundancy is weakened.","section":"Section 3.3.1, Eq. (4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The speedup numbers are impressive but conflate two orthogonal mechanisms. The average 18.67x speedup is achieved by FG Uni, which is the most aggressive variant. However, FG Uni also has the largest accuracy drop (-3.4 pp average). The paper should more clearly separate the contributions: SVTP alone gives ~2.78x (from Table 1, w/o SGAS row), SGAS alone gives up to 7.39x (w/o SVTP FG Uni). The product of these is ~20.5x, but the actual FG Uni achieves 18.67x, suggesting some interaction or overhead. The paper should discuss this discrepancy.","section":"Section 4.2, Table 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The SGAS thresholds (Table 5) were 'set from the output statistics of 100 LLaDA-V samples on MMMU' (Appendix B.5). This means the thresholds are tuned on one benchmark and then applied to others. While this is a reasonable practice, it means the method is not entirely parameter-free; the thresholds are a free parameter. The paper should discuss sensitivity to these thresholds. More importantly, the held-out TextVQA result (Table 10) shows that these thresholds, tuned on MMMU, do not generalize well to text-dense inputs. This reinforces the concern about the surrogate's validity.","section":"Section 4.1, Appendix B.5","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The abstract states 'an average 18.67 times inference speedup across six vision-language benchmarks without training, architectural modification, or attention caching, while maintaining competitive accuracy.' The phrase 'competitive accuracy' is vague. Consider quantifying the accuracy retention (e.g., 'within 3.4 pp of the baseline').","section":"Abstract, line 6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The table uses 'Best 2nd best' notation but the coloring is not visible in the text. Consider using bold or other markers. Also, the 'w/o SGAS' row uses 32 denoising steps, but its speedup is normalized by ÷2 to discount the inherent gain from halving the step count. This normalization is unusual and should be explained more clearly in the caption.","section":"Table 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The Pareto frontier figure is informative but the 'band' for the frontier is not clearly defined. Is it the convex hull? The caption should explain how the frontier is computed.","section":"Figure 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The notation '⊮' for indicator function is non-standard. Consider using '𝟙' or 'I'. Also, the definition of 'E' as 'granularity-controlled evaluation set' is unclear; it is later defined as the set of evaluation steps. Clarify.","section":"Section 3.3.2, Eq. (5)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The derivation of the output-stationarity bound is appreciated, but it relies on the assumption that 'the converged attention assigns negligible aggregate weight' to newly unmasked tokens. This assumption should be stated explicitly and justified, as it is critical for the bound to hold.","section":"Appendix D","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The held-out evaluation on POPE and TextVQA is valuable. However, the POPE results show that SGAS alone is lossless (88.30 F1) but adding SVTP costs 2.95 F1. This suggests that SVTP, not SGAS, is the main source of accuracy loss on object-existence tasks. The paper should discuss this trade-off more explicitly.","section":"Table 10","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the term 'competitive accuracy' without a formal definition. Given that the accuracy drops are non-negligible on some benchmarks (e.g., MM-Star -4.3 pp, MMBench -3.8 pp), the authors should state explicitly what they consider 'competitive' (e.g., within 5% of baseline).","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The headline result—18.67× wall-clock speedup on LLaDA-V with code released—is the real thing. The authors systematically measure attention redundancy along two orthogonal axes (spatial concentration on a few visual patches, temporal stabilization after a formation phase) and build two complementary mechanisms that target each axis independently. The spatial pruning (SVTP) is simple and effective: score patches by question similarity once, keep one patch plus the base crop. The temporal early stopping (SGAS) uses lexical coherence and parseability as a stop predicate. The speedup numbers are measured, not fitted, and the ablation study cleanly separates the contributions of each component. The paper also evaluates on a second backbone (LaViDa) and compares against concurrent methods (RedVTP, D3ToM), which is good practice.\n\nThe soft spots are real but bounded. First, the SGAS stop predicate treats fluency as a sufficiency surrogate for correctness. The authors' own TextVQA result (64.6 → 52.5) shows this breaks on text-dense inputs, and they acknowledge it in Appendix A.1. This doesn't invalidate the speedup claim, but it means the \"competitive accuracy\" claim is conditioned on the predicate holding, and the paper should be clearer about where it doesn't. Second, the MMBench FG Uni accuracy (69.0) is a 12.9 pp outlier relative to the −3.4 pp average drop. The paper doesn't explain this. Third, the SGAS thresholds were fit on 100 MMMU samples, so the temporal \"prediction\" is partly calibration. The authors disclose this in Appendix B.5, which is honest, but it qualifies the \"measurement-grounded\" framing. Fourth, all accuracies are single-run point estimates with no error bars. This is common in this subfield but still worth noting.\n\nWho is this for? Anyone working on efficient inference for diffusion LLMs, especially multimodal ones. The redundancy measurement methodology is reusable, and the speedup is practically meaningful (2.6 → 51.9 tokens/s). The paper deserves a serious referee. The main issues are addressable: clarify the scope of the quality-preservation claim, explain the MMBench outlier, and add error bars or significance tests.\n\nRecommendation: Accept with moderate revision. The core speedup claim is solid, but the accuracy claims need tighter qualification.","headline":"Measured speedup is real and repeatable; the quality-preservation claim is weaker than the abstract suggests, but the paper is still a solid systems contribution.","tokens_in":32538,"tokens_out":584,"duration_ms":7164},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"DRAMA speeds up multimodal dLLMs 18.67× by exploiting attention redundancy","keywords":["diffusion large language models","multimodal LLMs","inference acceleration","attention redundancy","visual token pruning","adaptive early stopping","training-free acceleration","spatial and temporal redundancy"],"falsifier":"On a held-out benchmark requiring precise numerical or factual accuracy (e.g., MathVQA or a domain-specific QA dataset), if SGAS stops early and produces a fluent-but-wrong answer at high frequency, the coherence-as-sufficiency premise would be falsified. The paper's own TextVQA result (accuracy dropping from 64.60 to 52.50 under FG Uni) already points in this direction.","tokens_in":32976,"feed_emoji":"⚡","tokens_out":1818,"duration_ms":15550,"temperature":0.7,"pith_summary":"The paper argues that multimodal diffusion large language models (dLLMs) compute massive amounts of redundant attention during inference, and that this redundancy can be systematically measured and exploited without any training or architectural changes. The authors identify two persistent redundancy structures: attention mass concentrates on a tiny fraction of visual tokens (spatial redundancy), and attention distributions stabilize rapidly after a few initial denoising steps (temporal redundancy). They design DRAMA, a framework with two complementary mechanisms that target these redundancies: Spatial-Redundancy-Aware Visual Token Pruning (SVTP) selects the most relevant image patches once before generation begins, and Sufficiency-Guided Adaptive Stopping (SGAS) terminates denoising early once the decoded text is coherent and answer-shaped. Together, these achieve an average 18.67× inference speedup across six standard vision-language benchmarks on the LLaDA-V model while maintaining competitive accuracy (within 3.4 percentage points of the full 64-step baseline). If this is correct, it establishes that the dominant computational cost of multimodal dLLM inference can be removed with a measurement-grounded, training-free approach, offering a strong acceleration baseline for this emerging model class.","feed_headline":"18.67× speedup for multimodal dLLMs by cutting redundant attention","feed_subtitle":"DRAMA prunes visual tokens once and stops denoising early, matching baseline accuracy on most benchmarks with no training.","key_machinery":"The key machinery is the systematic dissection of attention redundancy along two axes: spatial (how attention mass distributes over visual tokens, measured by cumulative coverage curves and the κ(0.9) metric indicating the minimum visual tokens needed for 90% attention coverage) and temporal (how attention distributions evolve across denoising steps, measured by inter-step Jensen-Shannon divergence and the redundancy complement R_l(t) = 1 - JSD/ln2). This analysis motivates two mechanisms: Spatial-Redundancy-Aware Visual Token Pruning (SVTP), which scores image patches by cross-modal similarity to the question text and retains only the most relevant patches once before denoising, reducing the key-value sequence length from N to |P|·M+|T|; and Sufficiency-Guided Adaptive Stopping (SGAS), which uses decoded-text lexical coherence statistics (diversity, peak frequency, consecutive repetition) plus task-schema parseability as a sufficiency surrogate to terminate denoising early once the output has stabilized. The paper also derives a Pinsker-type bound linking JSD collapse to output stationarity, justifying the use of decoded text as a proxy for attention convergence.","core_discovery":"The paper's central discovery is that attention in multimodal dLLMs exhibits two forms of redundancy that are both persistent and exploitable: (1) spatial redundancy, where the attention mass over visual tokens is concentrated on only a handful of image patches (the top 1% of visual key tokens capture ~47% of attention mass), and crucially this concentration is step-invariant, meaning a single selection made before denoising remains valid across all T steps; and (2) temporal redundancy, where inter-step attention divergence (measured by Jensen-Shannon divergence) rapidly drops near zero after an initial structure-formation phase (5-10 steps) and remains in a stable plateau where redundancy scores exceed 0.99 across all transformer layers, meaning subsequent denoising steps recompute near-identical attention at full O(T·L·N²) cost for diminishing informational return. The paper further shows that these two redundancy structures are orthogonal, enabling multiplicative acceleration gains when targeted simultaneously.","pith_inferences":["The paper's load-bearing premise is that lexical coherence and task-schema parseability are sufficient surrogates for output correctness. This premise likely breaks on tasks requiring precise numerical reasoning or multi-step verification, where a fluent wrong answer passes the gate. Held-out TextVQA results (accuracy dropping from 64.60 to 52.50 under FG Uni) already show this failure mode on tex","The spatial pruning strategy (SVTP) relies on the question text being available before denoising. This limits applicability to settings where the conditioning text is known upfront; in interactive or multi-turn scenarios where the question is refined during generation, the pruning would need to be re-done.","The temporal redundancy analysis (three-phase JSD structure) may generalize beyond multimodal dLLMs to text-only dLLMs and even to diffusion models in other modalities (e.g., image, video), suggesting that early stopping based on output stabilization could be a universal property of iterative denoising.","The 18.67× speedup is measured on a single backbone (LLaDA-V) with a specific generation regime (128 tokens, deterministic decoding). The actual speedup on other backbones (e.g., LaViDa, which already pools visual tokens) is much smaller (1.73-1.84×), indicating that the reported headline number is highly architecture-dependent and should not be generalized naively."],"forward_implications":["Multimodal dLLM inference can be accelerated by over an order of magnitude with no training or architectural changes, establishing a new baseline that uniform step reduction cannot match.","The spatial and temporal redundancy structures are orthogonal, so their gains multiply; this suggests future acceleration methods should target both axes independently.","The measurement-grounded approach (dissecting attention before designing mechanisms) can replace the current ad-hoc, visualization-inspired design of acceleration methods for this model class.","The SGAS stopping predicate, based on lexical coherence and task parseability, can serve as a general-purpose early-exit criterion for diffusion language models, not just multimodal ones.","The 18.67× speedup brings multimodal dLLM throughput (51.9 tokens/s on average) into a regime competitive with autoregressive models of similar scale, potentially making dLLMs practical for interactive applications."],"supporting_citations":[{"why":"Introduces LLaDA, the text-only diffusion LLM that serves as the foundation for the multimodal LLaDA-V backbone used throughout the paper.","marker":"[38]"},{"why":"Introduces LLaDA-V, the multimodal dLLM that is the primary evaluation backbone for all experiments.","marker":"[52]"},{"why":"Introduces LaViDa, the second multimodal dLLM backbone used for model architecture ablation experiments.","marker":"[28]"},{"why":"Supplies the Jensen-Shannon divergence metric and the approach for measuring inter-step attention redundancy, adapted in this paper from spatial to temporal comparison.","marker":"[39]"},{"why":"Identifies high inference latency due to iterative denoising as a fundamental challenge for dLLMs, which DRAMA aims to address.","marker":"[2]"},{"why":"RedVTP is the closest prior training-free acceleration method for multimodal dLLMs; DRAMA compares against it as a concurrent baseline.","marker":"[50]"},{"why":"D3ToM is a concurrent multimodal dLLM acceleration method; DRAMA compares against it as a concurrent baseline.","marker":"[7]"}],"fun_headline_variants":[],"cache_read_input_tokens":0,"weakest_assumption_plain":"The SGAS stop predicate treats decoded-text lexical coherence (diversity, peak frequency, consecutive repetition) plus task-schema parseability as a sufficient stand-in for output correctness, assuming that a fluent, answer-shaped sequence is necessarily a correct one.","fun_headline_variants_meta":{"raw":{"variants":[]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":7.9e-05,"raw_usage":{"cost":7.85512e-05,"is_byok":false,"cost_details":{"upstream_inference_cost":7.85512e-05,"upstream_inference_prompt_cost":7.81872e-05,"upstream_inference_completions_cost":3.64e-07},"total_tokens":1066,"prompt_tokens":1064,"completion_tokens":2,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":256,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":0}},"tokens_in":1064,"tokens_out":2,"cost_source":"openrouter_usage","duration_ms":3277,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":0,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-13T06:39:16.872767+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"On a held-out benchmark requiring precise numerical or factual accuracy (e.g., MathVQA or a domain-specific QA dataset), if SGAS stops early and produces a fluent-but-wrong answer at high frequency, the coherence-as-sufficiency premise would be falsified. The paper's own TextVQA result (accuracy dropping from 64.60 to 52.50 under FG Uni) already points in this direction.","supporting_citations":[{"cited_title":"Large Language Diffusion Models","cited_arxiv_id":"2502.09992","evidence_quote":"Introduces LLaDA, the text-only diffusion LLM that serves as the foundation for the multimodal LLaDA-V backbone used throughout the paper."},{"cited_title":"LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning","cited_arxiv_id":"2505.16933","evidence_quote":"Introduces LLaDA-V, the multimodal dLLM that is the primary evaluation backbone for all experiments."},{"cited_title":"LaViDa: A Large Diffusion Language Model for Multimodal Understanding","cited_arxiv_id":"2505.16839","evidence_quote":"Introduces LaViDa, the second multimodal dLLM backbone used for model architecture ablation experiments."},{"cited_title":"Efficient diffusion transformer with step-wise dynamic attention mediators","cited_arxiv_id":null,"evidence_quote":"Supplies the Jensen-Shannon divergence metric and the approach for measuring inter-step attention redundancy, adapted in this paper from spatial to temporal comparison."},{"cited_title":"Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models","cited_arxiv_id":"2503.09573","evidence_quote":"Identifies high inference latency due to iterative denoising as a fundamental challenge for dLLMs, which DRAMA aims to address."},{"cited_title":"Redvtp: Training- free acceleration of diffusion vision-language models inference via masked token-guided visual token pruning.arXiv preprint arXiv:2511.12428, 2025","cited_arxiv_id":null,"evidence_quote":"RedVTP is the closest prior training-free acceleration method for multimodal dLLMs; DRAMA compares against it as a concurrent baseline."},{"cited_title":"D3tom: Decider-guided dynamic token merging for accelerating diffusion mllms","cited_arxiv_id":null,"evidence_quote":"D3ToM is a concurrent multimodal dLLM acceleration method; DRAMA compares against it as a concurrent baseline."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}