What is the secret behind selecting layers from different models ? :)

#1
by Shamane - opened

Amazing work. I can see you have selected different layers from different models. Is it random?

thanks πŸ€—, I tried in this merge to be kinda symmetrical, choosing the middle layers of Astridboros and some layers of the beginning and end of Zephyr

Thanks. Any logic behind the order? Also did you have to fine-tune the merged model again ?

Thanks. Any logic behind the order?

Symmetry, this way the model has in the middle 6 layers of PAIXAI/Astrid-3B and 6 layers of jondurbin/airoboros-3b-3p0(which both came from Astridboros), to have a bit of the style of both(kinda human-like/RP).

Also did you have to fine-tune the merged model again ?

I recommend fine-tuning it if you want, cause swapping and inserting layers seems to confuse the model a bit.

Aryanne changed discussion status to closed

Thanks a lot

Sign up or log in to comment