Evaluation Pitfalls in Post-Training Quantization of Diffusion Transformers: An Empirical Audit
Frontiers in Computing and Intelligent Systems · 30.10.2026 · 18 (2) · 87-96
DOI: 10.54097/v6881m05 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Computing and Intelligent Systems, online erschienen am 30.10.2026. Verfasst von Zeming Jin. Die Arbeit nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Diffusion transformers are moving toward practical deployment, and post-training quantization (PTQ) has emerged as a standard, retraining-free compression route. The conventions used to evaluate PTQ for diffusion transformers—the Fréchet Inception Distance (FID) of images produced by a fixed sampling pipeline, with a cheap calibration-error statistic as a configuration-search proxy—are typically inherited from prior codebases rather than audited. Auditing this pipeline through a DiT-XL/2 (ImageNet 256×256) case study of 209 controlled experimental records, this paper identifies three systematic pitfalls. First, sampling-pipeline configuration: two latent defects in the legacy pipeline manufactured spurious method-level differences, and controlled A/B reproduction (deviation below 1%) shows that correcting them reverses earlier conclusions on channel equalization. Second, proxy-metric failure: 32 records sharing a bit-identical calibration error span an FID range of 98.74, and its rank correlation with FID is statistically indistinguishable from zero in every experiment family tested. Third, calibration-protocol sensitivity: under the corrected protocol, an error-aware dynamic-programming grouping strategy is statistically indistinguishable from a fixed-interval baseline (273.62±1.78 vs. 274.32±1.80), while group-wise equalization turns from apparently beneficial to harmful (273.62→313.23). The only method-level conclusion that survives the protocol correction—under three seeds, a single model, and the 8-bit weight-and-activation (W8A8) configuration studied—is that grouped quantization outperforms no grouping (388.96→274.32). This work closes with a corrected evaluation-protocol checklist.
Zitieren
APA
Jin, Z. (2026). Evaluation Pitfalls in Post-Training Quantization of Diffusion Transformers: An Empirical Audit. Frontiers in Computing and Intelligent Systems, 18(2), 87-96. https://doi.org/10.54097/v6881m05
Vancouver
Jin Z. Evaluation Pitfalls in Post-Training Quantization of Diffusion Transformers: An Empirical Audit. Frontiers in Computing and Intelligent Systems. 2026;18(2):87-96. doi:10.54097/v6881m05
BibTeX
@article{Jin2026,
author = {Jin, Zeming},
title = {Evaluation Pitfalls in Post-Training Quantization of Diffusion Transformers: An Empirical Audit},
journal = {Frontiers in Computing and Intelligent Systems},
year = {2026},
volume = {18},
number = {2},
pages = {87--96},
doi = {10.54097/v6881m05}
}
EUVIMED