Search for a command to run...
Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache