You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, have you guys considered adding a support for Mixture-of-Experts models?
They're usually quite hefty in terms of size and would be a great opportunity to have them offload parameters to CPU.
Hi, have you guys considered adding a support for Mixture-of-Experts models?
They're usually quite hefty in terms of size and would be a great opportunity to have them offload parameters to CPU.
Examples:
Switch Transformers (https://huggingface.co/google/switch-base-256)
NLLB (https://github.com/facebookresearch/fairseq/tree/nllb/)
The text was updated successfully, but these errors were encountered: