O(N log N) attention on consumer GPUs — no quadratic bottleneck.
Standard attention scales as O(N²). Wave Field routes information with FFT wave convolution on a field — O(N log N), long context on ordinary GPUs. Details in the posts below and on GitHub.