- Rating
- 3
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper asks whether continuous-state flow language models (FLMs) reason more efficiently than discrete diffusion language models, measuring sequence accuracy under matched denoising budgets on maze planning (Maze10/15/20) and Sudoku-Extreme, plus zero-shot transfer to Standard Sudoku. Theoretically, Theorem 1 shows the FLM ODE state accumulates a temporal convex superposition of candidate token sequences; Theorem 2 shows retained non-top-1 historical support can change prediction margins while the hard state stays fixed, with a tight Ωt/2 bound for any hard-state readout; Theorem 3 bounds the decoder-risk gap between continuous states and their hard projections. Empirically, at matched parameter counts FLM leads in the few-step regime: one-step sequence accuracy is 95.7/95.9/71.4 versus 94.0/52.6/35.6 for the best discrete baseline on Maze10/15/20, and on Maze15 at 64 steps FLM reaches the 95% target with an 88.25M model versus 139.00M for MDLM (36.5% fewer parameters). Mechanism studies (candidate-retention curves, intermediate-state interventions with Digit-soft and top-k controls, matched decoder probes) support the claim that retained non-top-1 mass is used by later denoising steps. On Sudoku the advantage is confined to 1–2 steps; beyond that, enhanced discrete samplers overtake FLM.
Critical strengths
Carefully controlled comparison. Architectures and parameter counts are matched per task and size (Tables 5, 9), step budgets are matched with an explicitly disclosed evaluation convention (K forward passes for FLM vs K+1 for discrete samplers), and the baseline suite includes exactly the methods that could break the claim (entropy/confidence remasking, ReMDM). The few-step advantage reproduces across three maze sizes, three model sizes per task, wall-clock latency (Table 10), and zero-shot transfer to Standard Sudoku (29.5 vs 27.1 at one step).
The theory–experiment loop is genuine. The superposition account (Theorems 1–3) is tested with interventions that include proper controls — Digit-soft isolates the projection operation itself, renormalized vs non-renormalized top-k separate rescaling from information removal, and an independent validation cohort reproduces the effect (47 of 48 correct-candidate events positive) — plus matched decoder probes whose largest Raw–Hard F1 gap occurs at step k=4 of an 8-step trajectory (0.8080 vs 0.5929). This is more than correlation-level evidence that retained candidate mass is computationally used.
Critical weaknesses
The headline comparisons carry no statistical support. Tables 1–2 appear to be single runs, and the test sets are 1,000 mazes per size and 1,000 Sudoku puzzles, where binomial noise alone is roughly ±1.5 points at p≈0.37. The Sudoku few-step margins the abstract leans on (21.0 vs 19.2 at one step, 36.7 vs 36.2 at two steps) lie within that noise, and no seeds, confidence intervals, or repeat runs are reported for any main-table entry.
The claimed advantage lives in a narrow regime on narrow tasks. On Sudoku, FLM plateaus after two steps and is clearly beaten from 16 steps onward (at 512 steps: 45.5 vs 62.2 for MDLM+Conf on Sudoku-Extreme); on Maze20 the 258.81M FLM never crosses 95% (92.9 at 512 steps); and all tasks are character-level synthetic outputs with no text-reasoning or language-modeling check, so whether "continuous states reason more efficiently" extends beyond toy planning remains open.
Questions for authors
How many seeds or runs stand behind Tables 1 and 2? In particular, are the 1–2 step Sudoku gaps (21.0 vs 19.2; 36.7 vs 36.2) stable across seeds given the 1,000-example test sets?
What is the rationale for the unequal maze training budgets (1,000 epochs for FLM vs 2,000 for discrete baselines, Table 6)? Does the few-step ordering change under equal epochs or equal wall-clock training compute?
Under strictly matched model evaluations (FLM at 2 steps vs a discrete sampler at 1 step, both = 2 evaluations), does the one-step advantage persist? The current convention grants discrete samplers one extra evaluation at every budget.
Is there any evidence the effect survives outside character-level synthetic tasks, e.g. a small text-reasoning benchmark or general language-modeling perplexity check?
3 comments
Response to Q3 and Q4 (Part 2/2)
[Q3 Strictly matched model evaluation]
We report the results under strictly matched numbers of model evaluations (NFE) below. FLM retains its advantage in the low-NFE regime. We have also included these results in Appendix D.8 (p. 32) of the revised paper.
| Dataset | Method | NFE=2 | NFE=4 | NFE=8 | NFE=16 | NFE=32 | NFE=64 |
|---|---|---|---|---|---|---|---|
| Maze20 | FLM | 88.1 | 91.3 | 92.1 | 92.4 | 92.5 | 92.5 |
| Maze20 | MDLM | 24.0 | 29.7 | 35.8 | 42.6 | 51.3 | 63.0 |
| Maze20 | MDLM + Ent. | 24.0 | 22.0 | 22.6 | 22.6 | 23.1 | 23.2 |
| Maze20 | MDLM + Conf. | 24.0 | 31.2 | 34.2 | 39.4 | 45.8 | 55.2 |
| Maze20 | SEDD | 32.9 | 40.6 | 47.3 | 55.0 | 64.8 | 73.9 |
| Maze20 | SEDD + Ent. | 32.9 | 58.0 | 82.8 | 90.2 | 90.8 | 88.5 |
| Maze20 | SEDD + Conf. | 32.9 | 62.1 | 71.1 | 76.7 | 82.3 | 84.4 |
| Maze20 | D3PM | 26.8 | 32.9 | 39.9 | 48.0 | 57.8 | 68.9 |
| Maze20 | D3PM + Ent. | 26.8 | 27.2 | 28.0 | 28.0 | 28.5 | 28.0 |
| Maze20 | D3PM + Conf. | 26.8 | 50.5 | 55.4 | 62.9 | 70.5 | 79.9 |
| Sudoku-Extreme | FLM | 36.7 | 44.2 | 45.2 | 46.3 | 44.7 | 43.9 |
| Sudoku-Extreme | MDLM | 20.3 | 24.9 | 31.0 | 39.5 | 44.7 | 46.7 |
| Sudoku-Extreme | MDLM + Ent. | 20.3 | 42.4 | 48.4 | 49.9 | 53.4 | 54.2 |
| Sudoku-Extreme | MDLM + Conf. | 20.3 | 28.0 | 36.6 | 41.9 | 49.5 | 54.0 |
| Sudoku-Extreme | SEDD | 18.8 | 24.0 | 30.2 | 38.0 | 44.7 | 45.7 |
| Sudoku-Extreme | SEDD + Ent. | 18.8 | 43.4 | 48.9 | 53.4 | 53.4 | 53.9 |
| Sudoku-Extreme | SEDD + Conf. | 18.8 | 26.3 | 33.7 | 39.5 | 45.7 | 50.4 |
| Sudoku-Extreme | D3PM | 11.6 | 16.5 | 24.3 | 31.4 | 38.7 | 41.4 |
| Sudoku-Extreme | D3PM + Ent. | 11.6 | 30.9 | 40.0 | 41.8 | 42.4 | 46.6 |
| Sudoku-Extreme | D3PM + Conf. | 11.6 | 19.5 | 30.5 | 36.7 | 41.2 | 48.1 |
[W2 & Q4 Scope and generalizability]
This paper primarily focuses on structured reasoning tasks, particularly planning and logical reasoning. These settings allow us to perform targeted interventions on intermediate states to further investigate the underlying mechanism. We consider extending our study to natural-language reasoning, including math and coding tasks, as future work. Moreover, our key strength is providing a theoretical explanation for FLM’s reasoning advantage with few inference steps.
We have made this scope more explicit throughout the paper to clarify the scope of our findings. Specifically, we have revised the abstract (p. 1, Lines 11--14), introduction (p. 1, Lines 47--48), and conclusion (p. 10, Lines 533--534) to emphasize our focus on structured reasoning and identify extensions to natural-language reasoning as future work.
We sincerely hope that we have addressed the concerns of the reviewer satisfactorily in the revised version and would kindly ask the reviewer to update their score accordingly.
Dear Reviewer,
As the discussion period draws to a close, we hope that our responses and revised manuscript have satisfactorily addressed your concerns. We would appreciate any further feedback and would be happy to discuss any remaining questions.
Thank you for your time and consideration.
Response to Q1 and Q2 (Part 1/2)
We thank the reviewer for recognizing our carefully controlled comparisons and the genuine interplay between theory and experiments. We also appreciate the reviewer’s valuable feedback, which has helped us improve the quality of our paper.
We have uploaded a revised version of the paper, with all changes highlighted in blue. All references to pages, sections, figures, tables, and equations use the revised version's numbering.
[W1 & Q1 Results across multiple seeds]
Our original experiments used seed 0, and we added Sudoku experiments with seeds 1 and 2. The two tables below report the mean results across all three seeds, showing FLM’s consistent reasoning advantage with few denoising steps. These results are also included in the revised paper’s appendix.
Sudoku-Extreme: mean ± sample standard deviation (%).
Standard Sudoku: mean ± sample standard deviation (%).
[Q2 Training-budget fairness]
We use different training durations because discrete methods converge more slowly. While FLM has converged by 1,000 epochs, the discrete baselines have not, so we extend their training to 2,000 epochs until convergence. Additional results for MDLM, SEDD, and D3PM at epoch 1,000 on Maze 20 are shown below.