SyncAI.news, a Varaisys broadcasting
HeadGuard: Selective Head Protection for Low-Bit VLM KV-Cache Quantization
NB

Nenad Banfic

· 1 min read

ResearcharXiv cs.CV

HeadGuard: Selective Head Protection for Low-Bit VLM KV-Cache Quantization

arXiv:2609.35800v1 Announce Type: cross Abstract: Low-bit key-value (KV) cache quantization saves storage but can sharply degrade vision-language model (VLM) accuracy. We introduce HeadGuard, a composable head-protection method that augments a base KV-cache quantizer with a fixed high-precision mask. Image-sensitivity and output-sensitivity scores select physical KV heads offline, with approximately 1/8 protected in the main experiments; their image keys and optionally values remain in bfloat16 (BF16), while the base quantizes unprotected image entries. Across eight VLMs, three base quantizers, and eight benchmarks (six discriminative and two generative), HeadGuard recovers a substantial fraction of lost accuracy on weaker quantizers, with the strongest gains for Qwen and InternVL. At 2 bits, the six-task discriminative mean over eight models rises from 0.436 to 0.580 on the weakest base; protection can also improve generated answers and caption fidelity to BF16 outputs. Mean accuracy gains persist across all three quantizers with both tested calibration datasets. Keys-only protection retains substantial recovery at lower modeled storage cost. Evaluated through simulated quantization, HeadGuard offers a composable way to improve low-bit VLM accuracy without replacing the underlying quantizer.

Original source

This story was published by arXiv cs.CV and written by Nenad Banfic. SyncAI.news shows a preview; the complete article is on the publisher's site.

Read the full story on arxiv.org

Similar News