metadata
license: apache-2.0
EDIT: Base Mistral, but with some minor head trauma. Promising in theory, needs finetuning, but only really outperforms the 14b in size.
A 11b Mistral base model, based on the NeverSleep recipe.
Recipe
slices
sources:
- model: mistralai/Mistral-7B-v0.1
- layer_range: [0, 24]
sources:
- model: mistralai/Mistral-7B-v0.1
- layer_range: [8, 32]
merge_method: passthrough
dtype: bfloat16