Head-Aware KV Cache Compression for Efficient Visual Autoregressive Modeling

VLM
Re-assign community