Positive users appreciate Fixed-Point Flow Maps for keeping the benefits of self-conditioning in one-step language models without architectural debt, while negative users criticize self-conditioning as a clumsy workaround.
Based on 3 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@sedielem so they found a way to keep the good parts without the architectural debt
@sedielem self conditioning always felt like such a clumsy workaround
@sedielem looks great I'll need to read this
It's always bothered me that self-conditioning is seemingly necessary to get good results with continuous language diffusion: it breaks the statelessness on which so much powerful machinery is built (e.g. distillation). This paper goes a long way towards cleaning up that mess! https://twitter.com/wognsfjq96/status/2072496642411147271
@sedielem so they found a way to keep the good parts without the architectural debt
@sedielem self conditioning always felt like such a clumsy workaround
It's always bothered me that self-conditioning is seemingly necessary to get good results with continuous language diffusion: it breaks the statelessness on which so much powerful machinery is built (e.g. distillation). This paper goes a long way towards cleaning up that mess! https://twitter.com/wognsfjq96/status/2072496642411147271
Positive users appreciate Fixed-Point Flow Maps for keeping the benefits of self-conditioning in one-step language models without architectural debt, while negative users criticize self-conditioning as a clumsy workaround.
Based on 3 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.