Release weights of smaller Experimental MoE

#12
by shahules786 - opened

Thanks a lot for your wonderful work.
The blog here mentions some smaller models that were trained for experimental purposes - namely DBRX-A (7B) and DBRX-B (23B), would it be possible to release the weights for those models too?

I belive that this would help the OSS community to experiment more in GPU-poor situations.

Thank you.

+1 would love to see the smaller models released. The 23B variant is a size the open source community really needs. Thank you very much for considering this.

Databricks org

Interesting suggestion, cc: @jfrankle

Sign up or log in to comment