Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache

Xiaol.x - Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache