- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper proposes CALP, which predicts response length from frozen masked diffusion model representations using a lightweight MLP. The predicted length determines a fixed canvas before denoising. Experiments on three backbones and five benchmarks report improvements in quality and inference efficiency.
Critical strengths
- The method is simple and leaves the backbone and denoising procedure unchanged, making integration straightforward.
- Evaluation across three backbones and five benchmarks provides useful evidence of practical benefits, with prediction overhead included in timing.
Critical weaknesses
-
Upfront prediction is insufficiently justified under block diffusion. Incremental block-wise extension with EOS stopping reduces the need to predict total length beforehand. Although CALP compares against dynamic methods, it does not isolate its benefit under matched execution and stopping conditions. Under-allocation can constrain responses, while over-allocation wastes computation and may impair termination.
-
The length model is oversimplified. Response length is naturally described by a conditional distribution , since the same prompt can admit multiple valid response lengths. CALP reduces this uncertainty to a single log-length estimate, without accounting for tail risk or asymmetric allocation costs. Distributional or conditional quantile prediction would better support uncertainty-aware canvas allocation.
Questions for authors
-
Does CALP-based allocation with full-canvas decoding offer a better trade-off in peak GPU memory, generation quality, and latency than block-wise generation with EOS stopping under comparable settings?
-
How accurately does CALP model the conditional response-length distribution, particularly its long tail? Do lower length-prediction errors consistently translate into better generation quality and efficiency?
-
Would CALP remain effective for chain-of-thought generation, where reasoning length can vary substantially for the same prompt? Would this require retraining the predictor or updating length estimates during decoding, and what fine-tuning data and procedure would be needed?
-
Can the authors track denoising trajectories under deliberately under- and overestimated canvases, including EOS emergence, truncation, and accuracy changes, to clarify how canvas allocation affects generation beyond computational cost?