- Rating
- 4(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
The paper proposes LiteCa, which uses low-rank AdaLN gate trajectories to aggregate per-module feature caches into a compact representation. It also introduces test-time correction to calibrate the aggregated components at full-computation steps.
Critical strengths
-
To the best of my knowledge, using low-rank AdaLN gate trajectories to construct compact, cross-module feature caches is a novel contribution to diffusion caching.
-
The method demonstrates substantial reductions in feature-cache memory across six image and video generation models, supporting its practical utility across different architectures.
Critical weaknesses
The gains over residual caching are not fully isolated. The main Wan2.1–HiCache configuration uses rank 0, where neither the gate-derived components nor TTC contributes. The reported results are obtained using different configurations across settings rather than a consistent configuration.
Questions for authors
-
Could the authors investigate whether rank can be selected from the gate spectrum, for example using a fixed cumulative-energy threshold, rather than performance on the evaluation set? The existing spectra and rank sweeps could help assess this criterion.
-
Could the authors analyze why nonzero ranks help some model–host pairs but not others? For example, comparing Qwen-Image and Wan2.1 in terms of gate reconstruction error and gains over rank-0 residual caching may clarify when additional components are useful.
-
The current TTC analysis mainly demonstrates error reduction on FLUX, but does not explain how well coefficients fitted at an anchor transfer to subsequent skipped steps. Could the authors analyze TTC’s error reduction as a function of distance from the preceding anchor, using the existing per-step results?
8 comments
Response to W2: Rank 0 on Wan2.1 with HiCache
We sincerely thank the reviewer for their careful review and recognition of our work.
A key advantage of LiteCa is that it uses a single cache representation for models with different caching preferences (Figure 2(d)). Rank 0 is also part of LiteCa, corresponding to residual caching. With coefficients fixed at 1, increasing the rank moves the representation closer to module-level caching. This allows LiteCa to work with models that favor different caching methods without changing its framework.
On Wan2.1, residual caching already performs similarly to module-level caching. The rank-1 component is also nearly orthogonal to the rank-0 error (Table R3E in our response to Q2), suggesting that adding more components provides little benefit. Thus, LiteCa at rank 0 achieves the best trade-off on Wan2.1: it matches the host in PSNR and SSIM (23.81 dB / .844 vs. 23.77 dB / .838), while using 97.8% less cache and achieving a higher speedup (3.09× vs. 2.85×). Increasing to rank 1 doubles the cache size without improving quality (23.76 dB / .842; Table 9). For models that benefit from additional gate components, Table R3A shows their contributions.
Response to W3: One configuration for all settings
We thank the reviewer for the constructive feedback.
We also evaluate LiteCa using a single configuration: rank 1 for every model and host (Table R3B; Appendix F, Tables 8–9). With this setting, LiteCa reduces cache memory by over 95% and runs faster than the host in all 10 model–host pairs. It also improves SSIM in all 10 pairs and PSNR in 9.
Table R3B. LiteCa with rank 1 across all model–host pairs, using the same schedules as Tables 1–2. PSNR↑ / SSIM↑ / LPIPS↓ are measured against full-step inference.
| Model | Host | Host | LiteCa (rank 1) | Cache reduction | Speedup (host → LiteCa) |
|---|---|---|---|---|---|
| FLUX.1-dev | TaylorSeer | 16.03 / .650 / .390 | 16.92 / .680 / .362 | 97.7% | 3.83× → 4.88× |
| FLUX.1-dev | HiCache | 16.85 / .677 / .348 | 17.25 / .692 / .354 | 97.7% | 3.81× → 4.88× |
| Qwen-Image | TaylorSeer | 14.66 / .556 / .424 | 17.83 / .608 / .319 | 98.3% | 2.73× → 3.20× |
| Qwen-Image | HiCache | 18.06 / .650 / .284 | 20.02 / .753 / .249 | 98.3% | 2.43× → 3.19× |
| HunyuanVideo | TaylorSeer | 21.00 / .744 / .219 | 22.07 / .772 / .198 | 98.4% | 3.12× → 3.46× |
| HunyuanVideo | HiCache | 22.12 / .763 / .197 | 22.81 / .783 / .193 | 98.4% | 3.21× → 3.45× |
| Wan2.1 | TaylorSeer | 21.93 / .808 / .133 | 21.99 / .812 / .132 | 95.6% | 3.01× → 3.07× |
| Wan2.1 | HiCache | 23.77 / .838 / .106 | 23.76 / .842 / .108 | 95.6% | 2.85× → 3.07× |
| MiniMax-H3 | TaylorSeer | 19.52 / .643 / .245 | 20.04 / .654 / .240 | 98.0% | 2.75× → 3.00× |
| MiniMax-H3 | HiCache | 20.33 / .664 / .225 | 20.87 / .676 / .223 | 98.0% | 2.86× → 3.02× |
Response to W1: Isolating the gain over residual caching
We thank the reviewer for the constructive feedback.
Table R3A breaks down the transition from residual caching to LiteCa into two stages: adding one gate component and applying TTC. With TTC enabled, LiteCa outperforms both residual caching and the module-level host in PSNR and SSIM across all four settings.
Table R3A. DrawBench-200, N = 6. Rank 0 is residual caching under the host’s own schedule and predictor. PSNR↑ / SSIM↑ / LPIPS↓.
Gate component (rank 0 → rank 1, TTC off). Adding one gate component brings LiteCa closer to module-level caching. This increases PSNR by 3.0 dB on Qwen-Image with TaylorSeer, where module-level caching outperforms residual caching, but decreases PSNR by 0.2 dB on FLUX.1-dev, where residual caching performs better. This model-dependent behavior, also shown in Figure 2(d), highlights the importance of a unified representation that spans both caching granularities.
Test-Time Correction (TTC off → on). With the same rank-1 cache, enabling TTC improves both PSNR and SSIM across both models and both host methods. Under TaylorSeer, TTC increases PSNR by 0.9 dB on FLUX.1-dev and 3.1 dB on Qwen-Image. Additional comparisons across N=3,6,9 are provided in Table G1 of our general response.