HeatKV doubles KV-cache compression ratios over prior methods for VAR models by creating static head-specific pruning schedules from attention rankings on a calibration set, while preserving image quality on Infinity-2B.
Simplifying representations, models, and datasets for efficient visual learning
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2representative citing papers
WAND converts full-attention AR-TTS models to global-plus-sliding-window attention with curriculum fine-tuning and teacher distillation, claiming quality preservation with up to 66.2% KV-cache savings and near-constant latency.
citing papers explorer
-
HeatKV: Head-tuned KV-cache Compression for Visual Autoregressive Modeling
HeatKV doubles KV-cache compression ratios over prior methods for VAR models by creating static head-specific pruning schedules from attention rankings on a calibration set, while preserving image quality on Infinity-2B.
-
WAND: Windowed Attention and Knowledge Distillation for Efficient Autoregressive Text-to-Speech Models
WAND converts full-attention AR-TTS models to global-plus-sliding-window attention with curriculum fine-tuning and teacher distillation, claiming quality preservation with up to 66.2% KV-cache savings and near-constant latency.