An Input-Dependent Readout Phase for Resonant Sequence Models

ResonatorLM uses a fixed learned phase per head to read a bank of damped complex oscillators, and making that readout phase a function of the current token adds input dependence to the read while keeping the convolution exact and costing 0.2 percent more parameters. In 6M-parameter character-level models on WikiText-2, the input-dependent phase matches the original at the training length and reduces perplexity from 12.81 to 3.76 when trained on 256 tokens and tested on 1024.