Writing study: Towards Sub-Second Molecular Docking as a Structural Primitive: A Quantized Consistency Diffusion Framework
← ICML 2026 oral writing corpus
Kexin Zhang, Weichen Qin, Yue Teng, Jiale Yu, Yuanyuan Ma, Jinyu Lin, Liping Sun, Jie Zheng, Jingyi Yu · Official ICML 2026 oral record
Figure, table and section ledger
Editorial contribution type: Efficiency, inference & systems (primary).
Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.
- Canonical PDF
- 30 physical pages
- Before references / appendix
- ≈ 9.67 page-equivalents
- Main figures / tables
- 3 / 1
- Attached appendix figures / tables
- 0 / 3
- Attached appendix span
- ≈ 16.91 page-equivalents
Source-bound correction: Source-checked physical-page text restores wrapped top-level heading labels on pp. 5. Heading starts and measured positions are unchanged; this is label completeness, not a new section. Newly detected unnumbered impact/singular-acknowledgment statements are retained from the size-aware source extraction; authored heading corrections are unchanged.
Section-by-section lengths and source pages
The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.
| Top-level heading | Starts on PDF page | Estimated span | Share of entire PDF |
|---|---|---|---|
| 1. Introduction | 1 | ≈ 1.42 pp. | 4.73% |
| 2. Related Work | 3 | ≈ 1.06 pp. | 3.53% |
| 3. Preliminaries | 4 | ≈ 1.34 pp. | 4.47% |
| 4. Methods: Accuracy–Latency Co-design for Real-Time Docking | 5 | ≈ 1.66 pp. | 5.53% |
| 5. Experimental Analysis | 7 | ≈ 2.17 pp. | 7.23% |
| 6. Contributions | 9 | ≈ 0.67 pp. | 2.23% |
| Acknowledgements | 10 | ≈ 0.15 pp. | 0.5% |
| Impact Statement | 10 | ≈ 0.47 pp. | 1.57% |
| References | 10 | ≈ 3.42 pp. | 11.4% |
| A. Unifying Diffusion and Transport Parameterizations | 14 | ≈ 5.57 pp. | 18.57% |
| B. Implementation Details: Featurization, Architecture, and Objectives | 19 | ≈ 4.81 pp. | 16.03% |
| C. Architecture Configuration, Training, and Inference Details | 24 | ≈ 1.14 pp. | 3.8% |
| D. Exploratory Experiments | 25 | ≈ 2.48 pp. | 8.27% |
| E. Limitations | 28 | ≈ 2.91 pp. | 9.7% |
All detected figure and table locations
Attached appendix
Argument and organization
- Problem and contribution
Treats input-information strength and call latency as separate evaluation axes, not as incidental implementation details behind a docking-accuracy headline.
- Argument sequence
A shared noise-ratio view motivates few-step dynamics; consistency tuning and selective quantization are then evaluated for both global docking and local geometry.
- Evidence and verification
The 0.17-second point generates five conformations for a 256-token complex on one H20 and excludes preprocessing, ranking and compilation; quality evaluation uses a different sample-and-rank protocol.
Problem, claim and contribution
Treats input-information strength and call latency as separate evaluation axes, not as incidental implementation details behind a docking-accuracy headline.
Introduction, p. 2 · Main paper
Argument and section order
A shared noise-ratio view motivates few-step dynamics; consistency tuning and selective quantization are then evaluated for both global docking and local geometry.
Sections 3-5, pp. 4-9 · Main paper
Figures, captions and visual explanation
Accuracy tables explicitly label the strong-prior regime, and four matched NFE/precision panels expose torsion degradation; the latency caption states hardware, output count, warmup and excluded stages.
Table 1/Figure 2, p. 8; Figure 3, p. 9; both PNGs inspected · Main paper
Tables, comparisons and evidence
The 0.17-second point generates five conformations for a 256-token complex on one H20 and excludes preprocessing, ranking and compilation; quality evaluation uses a different sample-and-rank protocol.
Section 5.5, p. 9; Appendix C, p. 25 · Main paper
Appendix and reproducibility
A four-level structural-prior experiment quantifies how information changes difficulty; limitations name non-linear length scaling and local geometry sensitivity under aggressive quantization.
Appendices D.2/E, pp. 26-28; Table 4, p. 30 · Main paper
Clarity and writing conventions
Protocol caveats are repeated beside the strongest numbers, although the end-to-end latency heading is broader than the timed GPU region actually measured.
Sections 5.3-5.5, pp. 8-9 · Main paper
What to reuse—and what not to infer
Transferable lesson: Attach every speed claim to its timed boundary and operating point, and expose unequal input information before comparing task accuracy.
Boundary: Do not combine the best accuracy and fastest latency into one unmeasured operating point, call the timing full-pipeline latency, or interpret docking benchmarks as prospective experimental validation.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-07
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-focused distributed manuscript scan, not exhaustive technical/proof validation. Sections 1-6, pp. 1-9: accuracy/latency and input-prior framing, generative parameterization, few-step consistency, mixed precision, task metrics and timed deployment configuration.
- Appendix / supplement
- Writing-focused appendix scan; code, data and external assets not executed. Appendices A-E, pp. 14-30: parameterization equivalences, implementation/training, exploratory controls, limitations and input-information hierarchy.
- Visually inspected PDF pages
- 8, 9
This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.