Metraipolis › archive › dc_77ef2c3abe03

LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization — arXiv:2610.03959

Source
arXiv
Original
https://arxiv.org/abs/2610.03959
License (as archived)
CC0-1.0
Archived
2026-10-06T05:24:11.716Z
SHA-256 of the archived text
ab835c2e68e75cca4b017a1565ce81b1de9669bb984f090835ee897144ce992a
Stable URL of this page
https://metraipolis.com/documento/dc_77ef2c3abe03

How to cite

LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization — arXiv:2610.03959 — arXiv, https://arxiv.org/abs/2610.03959 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_77ef2c3abe03 (Metraipolis archive dc_77ef2c3abe03, sha256 ab835c2e68e75cca4b017a1565ce81b1de9669bb984f090835ee897144ce992a)
BibTeX
@misc{dc77ef2c3abe03,
  title = {LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization — arXiv:2610.03959},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.03959},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_77ef2c3abe03; sha256 ab835c2e68e75cca4b017a1565ce81b1de9669bb984f090835ee897144ce992a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.03959. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recent world action models (WAMs) reuse pretrained video VAEs whose encoder latents directly condition downstream action policies. Quantization must therefore preserve not only reconstruction fidelity but also the policy-facing latent contract expected by the frozen policy. Direct NVFP4 leaves W4A4 quantization error uncompensated, whereas joint quantization-aware training (QAT) can recover reconstruction by moving this representation. On Wan2.1, joint QAT nearly matches FP32 VBench-7 (0.7403 versus 0.7409), yet LIBERO success collapses from 95.5% to 10.5%. Controlled decoder-only experiments show that activation quantize-dequantize operations alter the reconstruction signal and decoder Jacobian, redirecting the gradient returned to the encoder and inducing persistent latent drift. Based on this mechanism, we introduce LatentQuant, a two-stage NVFP4 QAT framework that first aligns the quantized encoder with its high-precision counterpart, then freezes it while adapting the decoder. Across Wan2.1 and Wan2.2, LatentQuant preserves near-baseline control and high reconstruction quality, achieving 95.75% success on LIBERO and 68.8% on RoboTwin. On NVIDIA B300 GPUs, NVFP4 execution achieves 1.17x-1.26x end-to-end VAE speedups over BF16 cuDNN.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.03959 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-06T05:24:05.410Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).