Fix diffusion head test time layer norm statistics

octo-models / octo

Octo is a transformer-based robot policy trained on a diverse mix of 800k robot trajectories.

https://octo-models.github.io/

MIT License

787 stars 152 forks source link

Closed mees closed 4 months ago

mees commented 4 months ago

Sets the dropout in the diffusion head to zero to avoid messing the layer norm with test time statistics