Question about the sampling process

[Ming-er]

Hi, Wenhao,
I am interested in your paper concerning latent flow matching for audio generation. However, I am a freshman in this field, and I am confused about the Euler sampling method implemented in lfm_scheduler.py, especially the purpose of calculating pred_sigma. I would appreciate it if you could provide more details or related reading materials for me. Thanks again!

[gwh22]

The calculation of pred_sigma aims to convert the model to diffusion model, but it is not required for the latent flow matching model, so the pred_sigma=pred. You can see the rectified flow paper and project for more details.