wrapping feature calculation in the inference_mode() instead of - Githubissues

facebookresearch / textlesslib

Library for Textless Spoken Language Processing

MIT License

518 stars 50 forks source link

wrapping feature calculation in the inference_mode() instead of #5

Closed eugene-kharitonov closed 2 years ago

eugene-kharitonov commented 2 years ago

no_grad()

Marginal speed-up on the probing task