Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache Paper • 2506.11886 • Published Jun 13, 2025 • 20 • 4