- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
Based on my understanding, this paper studies potential conflicts between tokens during parallel decoding in dLLMs. In the analysis part, it introduces valid answer uncertainty as a metric to quantify these conflicts. In the method part, it proposes AnF decoding, which works in three steps: selecting an anchor, refining the remaining predictions conditioned on that anchor, and then committing follower tokens in parallel. I find the research topic interesting, but I have concerns about the writing, the experimental setup, the method design and the evaluation results.
Critical strengths
S1. Personally, I find the topic of this paper interesting. However, I am not sure whether the broader community would see it as a key scientific problem, or whether the paper's novelty meets the expected bar. I will therefore take the other reviewers' opinions into account before making an overall assessment of novelty.
S2. The paper's entry point, token conflicts during parallel decoding, is interesting. To be honest, though, similar ideas exist in other works, albeit with a different framing.
S3. The method is relatively flexible.
Critical weaknesses
W1. The writing needs substantial improvement. A paper should convey its core motivation, key scientific question and research approach concisely, clearly and in an easy-to-follow way. At present, many statements in the paper, including in the abstract, introduction and method sections, are not sufficiently clear or rigorous.
W2. The reported results raise concerns about cherry-picking. I would like the authors to provide a more extensive and detailed hyperparameter sensitivity analysis, including a comparison of sensitivity across different models, to establish the robustness of the method.
W3. The anchor and follower concepts in this paper emphasize first committing an anchor during decoding so that it shapes the remaining context. These concepts are related to ADLM (https://openreview.net/pdf?id=E8adS5srds) and CoTA (https://openreview.net/pdf?id=mOz9jVYxsD), and the authors should discuss these works in the paper.
W4. The reliability of the valid answer uncertainty metric N_t still lacks supporting evidence. There is no rigorous mathematical proof of it, and the experiments lack statistical evidence. It would help to add a post-hoc statistical analysis of the actual conflicts that occur under parallel decoding, as corroborating evidence for N_t. Showing the correlation between these two statistics would make the metric considerably more convincing.
W5. The paper does not make clear the logical connection between its analysis and its proposed method. AnF also appears to perform poorly in Table 5. Moreover, the results in App. G.2 seem to contradict the paper's motivation. I therefore have doubts about the method's actual effectiveness, and I wonder whether the paper's narrative overstates what the method achieves.
W6. How common is cross-token inconsistency, and what drives it? Beyond the controlled tasks and illustrative examples such as Fig. 8, the paper does not show how common this phenomenon is on real benchmarks, so it is unclear whether the examples are representative. The authors should analyze two dimensions:
- Decoding steps. The effect may partly reflect model uncertainty in early decoding steps rather than inconsistency between tokens. The authors should break down conflicts by step, and separate tokens that are individually low-confidence from tokens that are each confident but jointly inconsistent.
- Position distance. The authors should report how far apart conflicting tokens typically are, and clarify how a "conflict" is defined for distant tokens.
Questions for authors
The evaluation should report the approximate caching and block settings in more detail.
5 comments
[W2] Robustness Across Settings and Models
Thank you for the detailed comment regarding robustness.
All methods use greedy decoding, and we evaluate each baseline over parameter ranges based on its original paper. We report accuracy–NFE trade-offs across the evaluated settings and the highest-accuracy operating point for each method, rather than selecting a single favorable configuration.
Beyond LLaDA-8B-Instruct, we evaluated two additional models, LLaDA-1.5 and Dream-Instruct-7B, and further added LLaDA2.0-mini, using the same hyperparameter settings as LLaDA-8B-Instruct without model-specific tuning.
[A for W2] Across different models and parameter settings, AnF generally achieves the highest or competitive accuracy among parallel decoders while maintaining competitive efficiency. These results demonstrate the robustness of AnF across models and decoding configurations.
Details
I. Evaluation Protocol and Parameter Sensitivity
All methods use greedy decoding and are evaluated using the same prompts and evaluation procedures. For each baseline, we follow the configurations proposed in its original paper and sweep the method-specific parameters over the ranges specified in Appendix E.
Rather than reporting results from a single configuration, we present accuracy–NFE curves across multiple parameter settings for each method, allowing a more comprehensive comparison of accuracy–efficiency trade-offs.
We analyze AnF's sensitivity to the follower threshold , target confidence , and candidate budget on LLaDA-8B-Instruct (Figure 21). Additional and evaluations on LLaDA-1.5 and Dream-Instruct-7B are presented in Figures 25 and 26. The generation-length comparison is provided in Table 8.
II. Robustness Across Models
Table A. LLaDA-1.5 (Table 9)
| Method | HumanEval | MBPP | GSM8K | IFEval | ||||
|---|---|---|---|---|---|---|---|---|
| Acc. | NFE | Acc. | NFE | Acc. | NFE | Acc. | NFE | |
| Sequential (top-1) | 45.7 | 512.0 | 48.2 | 512.0 | 83.7 | 512.0 | 64.9 | 1024.0 |
| Threshold | 26.8 | 32.4 | 43.2 | 21.6 | 49.3 | 25.0 | 47.0 | 120.8 |
| PUNT | 25.0 | 170.9 | 41.2 | 112.1 | 69.7 | 502.9 | 54.5 | 540.0 |
| DAPD | 40.2 | 55.1 | 47.5 | 106.8 | 78.2 | 79.5 | 63.0 | 72.0 |
| ETE | 34.8 | 121.8 | 40.1 | 29.9 | 81.9 | 121.0 | 59.3 | 259.7 |
| AnF | 46.3 | 79.2 | 49.8 | 29.1 | 78.5 | 78.2 | 63.6 | 211.9 |
Table B. Dream-Instruct-7B (Table 10)
| Method | HumanEval | MBPP | GSM8K | IFEval | ||||
|---|---|---|---|---|---|---|---|---|
| Acc. | NFE | Acc. | NFE | Acc. | NFE | Acc. | NFE | |
| Sequential (top-1) | 49.4 | 512.0 | 62.6 | 512.0 | 83.5 | 512.0 | 52.9 | 1024.0 |
| Threshold | 28.0 | 150.7 | 52.1 | 53.0 | 44.9 | 165.2 | 29.8 | 20.2 |
| PUNT | 23.2 | 203.9 | 44.4 | 99.0 | 57.2 | 151.2 | 48.4 | 162.9 |
| DAPD | 26.8 | 243.1 | 27.2 | 298.5 | 64.4 | 108.5 | 31.1 | 424.1 |
| ETE | 29.9 | 64.3 | 49.0 | 120.0 | 73.7 | 120.1 | 36.4 | 257.1 |
| AnF | 50.0 | 31.6 | 65.8 | 37.6 | 84.8 | 65.9 | 60.8 | 206.8 |
Table C. LLaDA2.0-mini (Table 11)
| Method | HumanEval | MBPP | GSM8K | IFEval | ||||
|---|---|---|---|---|---|---|---|---|
| Acc. | NFE | Acc. | NFE | Acc. | NFE | Acc. | NFE | |
| Sequential (baseline) | 73.2 | 228.7 | 77.0 | 70.5 | 90.1 | 276.0 | 82.6 | 360.0 |
| Threshold | 71.3 | 31.8 | 75.5 | 17.2 | 90.1 | 116.3 | 81.9 | 292.5 |
| PUNT | 60.4 | 161.4 | 66.5 | 78.3 | 88.9 | 399.3 | 80.6 | 703.0 |
| ETE | 57.9 | 25.5 | 60.3 | 10.4 | 88.6 | 79.8 | 76.2 | 176.7 |
| AnF | 72.6 | 39.0 | 75.1 | 17.9 | 91.0 | 108.9 | 81.5 | 282.6 |
Across the evaluated models, AnF achieves the highest accuracy among parallel decoders on most benchmarks and remains competitive on the others, while maintaining favorable accuracy–NFE trade-offs.
These results show the robustness of AnF across different diffusion language models.
[W3] Discussion of ADLM and CoTA
Thank you for pointing out these related works.
[A for W3] We added ADLM and CoTA to Related Work (Section 2), discussing training-based anchor prediction in ADLM and how context tokens guide subsequent predictions in CoTA.
[W4] Reliability of Valid Answer Uncertainty
Thank you for the detailed comment regarding the reliability of . In this revision, We strengthened its justification through both theoretical and empirical analyses of its relationship with answer-mixing risk.
We added an analysis of the conflict probability under independent parallel completion, deriving a theoretical bound relating $C_t$ to and examining their empirical relationship across four analysis tasks.
The analysis is presented in Section 4.3, with the definition and proof in Appendix C.
[A for W4] We establish , showing that low guarantees low conflict probability under independent parallel completion. Empirically, states with low exhibit low conflict probability across all four tasks.
Details
I. Conflict
To examine how relates to answer mixing, we compare it with the conflict probability . A completion is conflicting if it contains at least one individually valid unit, but no single valid answer contains all of those units. We define
measures the probability of answer mixing when all remaining masked positions are completed independently using the current predictions.
II. Relation between and Conflict Probability
Theoretical relation. Conflicting completions contribute no answer contribution mass, so . Since , we have . Combining these inequalities yields
Thus, a low guarantees a low conflict probability under independent parallel completion.
Empirical relation. We compare with , estimated from 512 independent completions at pre-commitment states of top-1, top-2, and top-4 decoding across four analysis tasks.
Table D. Correlation between and by range
| Task | |||
|---|---|---|---|
| Graph | 0.31 | 0.48 | 0.55 |
| Replace Random | 0.81 | 0.59 | 0.09 |
| Shuffle | 0.79 | 0.84 | 0.57 |
| Latin Square | 0.96 | 0.96 | 0.98 |
Table D shows positive correlations between and across all four tasks for , ranging from 0.48 to 0.96.
These results show that low is associated with low conflict probability, indicating that reflects answer-mixing risk in practice.
[W6] Frequency and Characteristics of Answer Mixing on Generation Benchmarks
Thank you for your comment, which helps improve the readability of our paper.
We added a direct analysis of answer mixing across four generation benchmarks. We then measure the frequency of answer mixing and analyze the confidence and distance of mixed tokens. The analysis is presented in Section 6.2 and Appendix F.
[A for W6] Answer mixing increases from 3.7% under top-2 to 5.3% under top-4, while AnF achieves 1.6% despite committing more tokens per step. Mixed tokens are typically close together, and mixing can occur even at high confidence, showing that token confidence alone does not prevent answer mixing.
Details
I. Measuring Answer Mixing
We use GPT-6.1 to identify mixed parts in generated outputs, distinguishing answer mixing from ordinary prediction errors. We then match these parts to decoding traces and count mixing only when the identified parts are committed in the same decoding step.
The judging protocol and full prompt are provided in Appendix F.
II. Frequency of Answer Mixing
Table E. Mixing rate and tokens committed per step
| Method | Mixing (%) | Tokens/step |
|---|---|---|
| Top-2 | 3.7 | 1.99 |
| Top-4 | 5.3 | 3.95 |
| Threshold | 2.6 | 2.51 |
| Fast-dLLM | 1.4 | 2.33 |
| PUNT | 6.0 | 6.88 |
| DAPD | 8.5 | 3.64 |
| ETE | 5.5 | 6.49 |
| AnF | 1.6 | 5.11 |
Mixing increases with parallelism from top-2 to top-4. In contrast, AnF maintains a low mixing rate while committing more tokens per step than top-4.
These results are reported in Table 1, with additional results in Appendix F.
III. Characteristics of Answer Mixing
We analyze the distance between mixed parts and their confidence at commitment across decoding methods.
Table F. Distance and confidence of mixed tokens
| Method | Median distance | Mean confidence |
|---|---|---|
| Top-2 | 2 | 0.63 |
| Top-4 | 2 | 0.57 |
| Threshold | 3 | 0.98 |
| Fast-dLLM | 2 | 0.97 |
| PUNT | 7 | 0.89 |
| DAPD | 20 | 0.92 |
| ETE | 3 | 0.80 |
| AnF | 3 | 0.98 |
These results show that answer mixing occurs even at high token confidence, indicating that confident individual predictions can still produce cross-token inconsistency when committed in parallel.
We appreciate the detailed and thoughtful feedback. We included additional comments to further improve our work.
Thank you for the feedback on the clarity of the writing. We revised the paper to clarify the motivation, central research question, and proposed approach. The revisions are highlighted in blue.
Details
I. Abstract and Introduction (Section 1)
We rewrote the abstract and introduction to clarify the core motivation, central research question, and proposed approach. We made explicit how answer mixing arises during parallel decoding and how this motivates using anchor-conditioned refinement to reduce mixing while retaining parallel efficiency.
II. Preliminaries (Section 3)
We clarified the meaning and interpretation of valid answer uncertainty Nt in Section 3.3.
III. Analysis (Section 4)
We added Section 4.3 to clarify the relationship between valid answer uncertainty Nt and conflict risk. We established a theoretical bound on conflict probability and empirically showed that low Nt is associated with low conflict risk across four analysis tasks.
IV. Method (Section 5)
We clarified the motivation for the two-stage anchor selection and added a theoretical connection between predictive entropy and Nt in Section 5.1.
V. Experiments (Section 6)
We added a direct evaluation of answer mixing across four generation benchmarks, reporting mixing rates alongside accuracy, GPU-forward throughput, and memory in Table 1.
We further analyze the distance and confidence of mixed tokens in Figure 8 to clarify how answer mixing occurs during parallel decoding.