CleverMommy-mix-20b / README.md
athirdpath's picture
Update README.md
701714a
metadata
license: cc-by-nc-4.0

An extended part of my effort to create Eileithyia-20B. This model is made by following the recipe below, inverting it, then SLERPing the models back together at 0.5, hopefully fusing the models into one block for use with Harmonia.

slices:

  • sources: - model: microsoft/Orca-2-13b

    layer_range: [0, 16]
  • sources: - model: athirdpath/Eileithyia-13B

    layer_range: [8, 24]
  • sources: - model: microsoft/Orca-2-13b

    layer_range: [17, 32]
  • sources: - model: athirdpath/Eileithyia-13B

    layer_range: [25, 40]

merge_method: passthrough

dtype: float16

Thanks to Undi95 for pioneering the recipe.