Download the code and result files Watch a decodeEvery figureHow to reproduce
The paper's abstract
Three clocks run on every decode, and only the first one decides the answer
A masked-diffusion VLM starts from a fully masked answer and fills it in over T forward passes. At every pass we read the answer the model would give if stopped there, the entropy at the answer position, and the full text. On LLaDA-V the three settle at very different times.
The paper asks four questions
How a masked-diffusion VLM answers, and where the paper measures it
The image is encoded once. The language model then fills in a fully masked answer over T passes. At each pass it predicts a word for every masked position, keeps the most confident one and masks the rest again. Press play to send one real question through the loop: the box under the diagram shows what the model would answer if it stopped at that pass.
Watch a real decode, pass by pass
Pick a question and press play. The box shows what the model would output if it stopped at that pass; the highlighted word is the answer slot. The chart is the entropy at the answer slot (log scale). The answer stops changing long before the entropy collapses, and the text keeps changing after that.
A few passes out of thirty-two, on seven backbones
Iteration utilisation U is the last pass at which the emitted answer changes, divided by the number of passes: the share of the schedule that can still alter the answer. It needs no extra computation.
The confidence arrives late, and it measures the text, not the image
Commit granularity, not refinement: a frozen hedge, and one deferred round
Short schedules commit many tokens per pass. When the first pass puts a hedge template at the answer slot ("There, there is a …"), a coarse first round freezes it in place. The answer-deferred round (ADR) keeps the answer span out of the first commit round; nothing else changes.
Endpoint confidence is schedule-fragile; one early read fixes it
If the confidence tracks the text, a longer schedule that writes more text should make it a worse error detector, even as accuracy rises. Reading the same entropy at a fixed early pass avoids that.
Five pre-specified repairs of the endpoint do not come close
The interventions were fixed before seeing the data. None approaches the free trajectory statistic as an error detector; on GQA three rank errors below chance.
A certified early exit, with the random-score control that exposes vacuous certificates
The exit threshold is certified with distribution-free bounds at a flip-risk budget α. The same procedure run on a random score shows how much of the saving any score would get; on POPE flips are rare and the certificate is largely vacuous, on GQA it is not.
Seven backbones, fresh pools, image noise and a prompt intervention
Every table of the paper and its supplement
Parsed from the paper's LaTeX at build time, so the numbers are the paper's numbers. Where the supplement explains how to read a table, its note comes with it.
Every figure of the paper and its supplement
Click a figure to enlarge it and read its caption.
All the code: experiments, analyzers, tables and figures
The data module recomputes every number from the per-example records; the experiment drivers wrote those records on one GPU each.
Every number traces to one of these files
The list was captured by running each of the paper's data, table and figure scripts and recording every file it opened. Under each file: the scripts that read it.
Rebuild the tables and figures from the archive
Download the archive, unzip it and run from its top folder (Python 3.9+ with numpy, scipy, matplotlib and pillow).
python3 papers/cvpr2027_dvlm/make_supp_tables.py # every generated table python3 papers/cvpr2027_dvlm/make_backbone_tables.py # the main-text and backbone tables python3 papers/cvpr2027_dvlm/make_supp_figures.py # the supplementary figures (and the line figures) python3 papers/cvpr2027_dvlm/supp_data.py # recompute the main cache from the per-example records python3 papers/cvpr2027_dvlm/supp_data_backbones.py # recompute the seven-backbone cache (about 1.5 minutes)
How this page was checked
Nothing on this page is typed by hand. A single build script reads the paper's LaTeX, its figures, its caches and the per-example records, and it refuses to publish if any of these checks fails:
Limitations
The paper's Limitations section, read from its LaTeX at build time.