add setting for model output clean prediction
add ability to be guided with a smaller worse model, from a paper by … …Tero Karras
add maybe fourier features before speech attention
Merge pull request #46 from lucidrains/freq-axis some groundwork to prepare for attending across frequency tokens, as …
patch hl gauss again
patch again
allow for using HL-Gauss in the duration predictor last layer https:/… …/arxiv.org/abs/2403.03950
add hyper connections and cite, default to 4 residual streams