Boundary-Protection W8A8 HiFloat8 Quantization for Large-Scale Text-to-Video Diffusion Transformers
This work addresses the challenge of quantizing large-scale video diffusion transformers for efficient deployment on Ascend NPUs, but the evaluation is limited to 5 prompts and the gains are incremental.
The paper proposes a boundary-protection post-training quantization (PTQ) method for Wan2.1-T2V-14B, a 14B text-to-video diffusion transformer, achieving W8A8 HiFloat8 quantization with no measurable accuracy loss on five VBench dimensions.
We present a post-training quantization (PTQ) approach for Wan2.1-T2V-14B, a 14-billion-parameter text-to-video diffusion transformer, targeting the W8A8 HiFloat8 (HiF8) format on Ascend 910B NPUs. A central challenge in quantizing video DiT models is the heterogeneous activation distribution across transformer blocks: boundary blocks (the first and last few blocks) exhibit fundamentally different statistical properties from middle blocks, making uniform quantization ineffective. We conduct a systematic per-block activation analysis across all 40 WanAttentionBlocks and use the findings to motivate a boundary-protection strategy that retains the first two and last three blocks in BF16 while quantizing the remaining 35 blocks with W8A8 HiF8. The proposed PTQ method matches or marginally exceeds the BF16 baseline on all five VBench dimensions evaluated, indicating no measurable accuracy loss within the 5-prompt evaluation set. An ablation study over four protection configurations confirms that full boundary protection yields the highest average VBench score, validating the data-driven block selection. We additionally investigate quantization-aware training (QAT) as a complementary fine-tuning stage and analyze the conditions under which it fails to outperform plain PTQ on single-card hardware.