FMLM★ Distills Self-Conditioning Into One-Step Language Generation

Super nice work! I really like the warm start idea for the FP computation as an interpretation of the self-conditioning. We recently released a work on Fixed-Point Masked Generative Modeling which could be relevant to your work: https://arxiv.org/abs/2605.31215 We also have a specific warm start technique for FP-MGM.

Glad to have worked on this with @wonjung_ki13109, @FEijkelboom, @chandavidlee, @nmboffi, @hongseu33, @jw9730

@andreamiele_ Thanks! We believe this structure is universal and can be adapted to many other generative/predictive systems.