- Rating
- 3
- Confidence
- 3(+1)
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
The paper proposes De-Anchoring, a verifier for diffusion language models. It selects key tokens, such as derived numbers, option letters, and code identifiers, and groups their repeated occurrences into anchor chains. It masks one chain at a time, together with the final answer where defined, and scores the original tokens with one forward pass through a frozen model. It averages these log-probability scores to rank responses without generating new text. Experiments cover response selection and correctness scoring on nine benchmarks with two models, plus an RL application on math tasks.
Critical strengths
-
The method is simple and needs no verifier training or external model. Each anchor chain takes one forward pass. Tables 3 and 4 support the choice of informative anchors and the need to mask all their occurrences. Table 6 also shows that checking just one chain retains much of the selection accuracy. This makes the method a useful way to reduce verification work.
-
The gains are consistent across models and tasks. In Table 1, De-Anchoring beats BMC in all 18 model–dataset settings. For example, Dream's HumanEval accuracy rises from 46.34% to 51.22%, and LLaDA's MBPP accuracy rises from 46.95% to 51.75%. Table 11 shows gains with another sampling setup, and Table 2 generally supports better separation of correct and incorrect responses. The evidence goes beyond a single benchmark or sampling choice.
Critical weaknesses
-
The BMC comparison needs more detail. Section 5.2 says the comparison is budget-matched, but the settings behind Tables 1 and 2 are not clear: the mask ratio, number of reconstructions, and steps per reconstruction. Table 6 also needs an explanation of how K and N_BMC determine the total forward passes. These details matter because better accuracy at lower verification cost is a main claim. The current description makes that claim hard to check.
-
The RL comparison changes more than the verifier. In Table 13, BMC gives correct responses a continuous score-based reward, with a coefficient that increases during training. De-Anchoring gives a fixed reward to the top half of the correct responses and uses a constant coefficient. The gains in Table 7 could therefore come partly from the reward rule or schedule. This weakens the claim that a better verification signal causes the RL improvement.
Questions for authors
-
For BMC in Tables 1, 2, and 6, what mask ratio, number of independent reconstructions per candidate, and denoising steps per reconstruction did you use? How were these values chosen? Were total forward passes per candidate matched between the methods? Please also explain how K and N_BMC map to this budget in Table 6.
-
Could you compare against random masking with the same mask count, answer masking, scoring rule, and total forward passes? Tables 3–5 support the design, but this control would help show how much the chain structure adds beyond the amount of context kept visible. Matching mask counts here would test the design, rather than match compute cost.
-
In the RL experiment, does De-Anchoring still outperform BMC when both use the same rule for turning scores into rewards and the same reward coefficient schedule?
4 comments
Response to Reviewer S6d4 [Part 2]
W2 & Q3. In the RL experiment, does De-Anchoring still outperform BMC when both use the same rule for turning scores into rewards and the same reward coefficient schedule?
We thank the reviewer for suggesting this important ablation. To directly address your concern, we ran a controlled comparison in which De-Anchoring and BMC differ only in how the score is computed. Every other rule is shared, including how scores are turned into rewards and the schedule of the reward coefficient. In addition to matching the rule and schedule, we ensure identical verification costs for BMC and De-Anchoring by allocating exactly forward passes to BMC per candidate, where is the number of forward passes used by De-Anchoring for that candidate.
Under this fully controlled setting, De-Anchoring still outperforms BMC on each of the three math tasks: 81.73 vs. 79.30 on GSM8K, 24.61 vs. 16.41 on Countdown, and 33.00 vs. 32.80 on MATH. On both Avg and Pooled Avg, De-Anchoring outperforms every BMC configuration, including settings with a larger verification budget (i.e., fixed 8 forwards). This indicates that De-Anchoring provides a more efficient and robust verification signal than BMC, in line with its strong Best-of-N and AUROC results reported in the paper.
| Reward | GSM8K | Countdown | MATH | Avg | Pooled Avg |
|---|---|---|---|---|---|
| LLaDA-8B-Instruct (no RL) | 78.24 | 18.36 | 31.60 | 42.73 | 59.61 |
| diffu-GRPO | 78.77 | 32.40 | 45.00 | 60.82 | |
| + BMC (B-rule, 8 forwards) | 79.61 | 22.27 | 31.80 | 44.56 | 61.02 |
| + BMC (D-rule, 8 forwards) | 79.23 | 33.80 | |||
| + BMC (D-rule, forwards) | 79.30 | 16.41 | 32.80 | 42.84 | 60.34 |
| + De-Anchoring (D-rule, forwards) | 81.73 | 24.61 | 46.45 | 62.94 |
Avg is the unweighted mean over the three benchmarks; Pooled Avg weights each benchmark by its number of test questions (GSM8K 1319, Countdown 256, MATH 500). B-rule and D-rule denote the BMC rule and the De-Anchoring rule, respectively. Each rule fixes both how scores are turned into rewards and the schedule of the reward coefficient.
Response to Reviewer S6d4 [Part 3]
Q2. Could you compare against random masking with the same mask count, answer masking, scoring rule, and total forward passes? Tables 3–5 support the design, but this control would help show how much the chain structure adds beyond the amount of context kept visible. Matching mask counts here would test the design, rather than match compute cost.
We thank the reviewer for suggesting this valuable ablation study. Random masking that controls for the number of masked tokens in each chain provides an opportunity to directly test the effectiveness of our design, as the reviewer suggested. We conducted this experiment, and the results further demonstrate the clear advantages of our anchor-chain structure.
| Dataset | Masking Method | BoN | AUROC |
|---|---|---|---|
| GSM8K | De-Anchoring (anchor chains) | 80.16 | 0.764 |
| GSM8K | Same-count random positions | 62.96 ± 1.71 | 0.544 ± 0.005 |
| MATH | De-Anchoring (anchor chains) | 58.33 | 0.710 |
| MATH | Same-count random positions | 51.07 ± 0.80 | 0.516 ± 0.017 |
| Countdown | De-Anchoring (anchor chains) | 58.02 | 0.688 |
| Countdown | Same-count random positions | 31.30 ± 2.25 | 0.455 ± 0.020 |
Anchor-chain masking vs. random masking of the same masking amount. LLaDA-8B-Instruct, problems whose candidate sets contain both correct and incorrect responses. Mean ± std over three seeds.
Following Section 5.1 of our paper, we consider queries whose candidate sets contain both correct and incorrect responses. We evaluated GSM8K, MATH, and Countdown on the same judgment set used in Tables 3–5 (LLaDA-8B-Instruct; random masking results are reported as mean ± standard deviation over three seeds). Specifically, we replace each anchor chain with an equal number of randomly selected token positions, ensuring that the computational cost, number of masked tokens, and scoring rule remain identical, with only the masked positions differing. The results demonstrate that the anchor-chain design structurally establishes a robust and reliable verification setting.
[1] Ruan et al. Reasoning on the Manifold: Bidirectional Consistency for Self-Verification in Diffusion Language Models. ICML 2026.
[2] Reimers and Gurevych. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. EMNLP 2019.
If you have any further questions, please let us know.
Thank you very much, Authors
Thank you for the clarification and additional experiments. The BMC settings and the explanation of the forward-pass budget address my concern about compute matching. The RL comparison with the same reward rule and schedule addresses the confounding issue. The random-masking experiment also provides clearer evidence for the anchor-chain design.
My main concerns have been addressed. I maintain my rating of 3, with greater confidence in the recommendation. Please include these details and additional results in the revised paper.
Response to Reviewer S6d4 [Part 1]
Dear Reviewer S6d4,
We sincerely appreciate your review with thoughtful comments. We have carefully considered each of your questions and provide detailed responses below. Please let us know if you have any further questions or concerns.
W1 & Q1. For BMC in Tables 1, 2, and 6, what mask ratio, number of independent reconstructions per candidate, and denoising steps per reconstruction did you use? How were these values chosen? Were total forward passes per candidate matched between the methods? Please also explain how K and N_BMC map to this budget in Table 6.
We thank the reviewer for this thoughtful observation. To improve the clarity of our paper, we have incorporated the requested details into Section 6.2 (page 9) of the revised manuscript, with the additions highlighted in green.
BMC uses a mask ratio of 0.9, following the default setting in the original paper. The two BMC hyperparameters highlighted by the reviewer,
the number of independent reconstructions per candidateandthe number of denoising steps per reconstructionare respectively denoted by NBMC and T; then, BMC spends NBMC×T forward passes per candidate. For a fair performance comparison, we matched this computational budget to De-Anchoring on a per-candidate basis. De-Anchoring verifies m=min(∣V∣,K) anchor chains per candidate, using one forward pass per chain, where K is the cap on verified chains (K=8 in Tables 1 and 2). Accordingly, BMC must also be allocated exactly m forward passes for the same candidate, i.e., NBMC×T=m. However, when m is prime (e.g., 5 or 7), setting NBMC>1 forces T=1, which the original BMC paper [1] reports to be a highly unfavorable setting.To mitigate this fairness issue as much as possible, we adopt two measures in Tables 1 and 2. First, we set NBMC=1 and T=m, ensuring that both methods use exactly the same total number of DLM forward passes for every candidate. Second, to reduce the potential disadvantage to BMC arising from variance due to random masking, we report its performance averaged over three masking seeds.
Meanwhile, Table 6 further explores BMC's potential by increasing NBMC. Starting from the per-candidate budget m, for candidates whose budget reaches the column-wise cap K (i.e., m=K), we reallocate the budget as NBMC×T with T=K/NBMC, while all other candidates keep NBMC=1 and T=m. This setting allows us to explore BMC's potential more thoroughly while maintaining the same number of DLM forward passes for De-Anchoring and BMC. Nevertheless, increasing NBMC yielded only marginal differences and still failed to outperform De-Anchoring.
Importantly, we remark that we exclude the forward passes of BMC's external sentence encoder [2] from the reported computational budget, as they are not directly comparable to DLMs forward passes. Thus, BMC incurs additional FLOPs from the external model in all settings reported in Tables 1, 2, and 6. This makes the computational budget comparison favorable to BMC and conservative for De-Anchoring. Nevertheless, the results show that BMC remains within 65.5–66.1 across all budget allocations, consistently underperforming De-Anchoring at every matched K. Remarkably, De-Anchoring with K=1 even outperforms BMC with K=8 across all reported configurations.