-
Notifications
You must be signed in to change notification settings - Fork 445
Open
Description
Description of the feature request:
In the slicing configs, the smallest config is for 1.91B which corresponds to the E2B model. Is it possible to make a smaller model,i.e., 0.9B or smaller(maybe with 26 layers)? If yes, please specify the optimal slicing config for that case.
Also, similar to text model, is it possible to slice (reduce the number of layers for )the audio encoder?
What problem are you trying to solve with this feature?
Deployment on resource-constrained mobile devices(with 4-6GB RAM) and web
Any other information you'd like to share?
No response
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels