README.md

metadata

license: apache-2.0

Weyaxi/OpenHermes-2.5-neural-chat-v3-3-Slerp

This is the model for Weyaxi/OpenHermes-2.5-neural-chat-v3-3-Slerp. I used mergekit to merge models.

Yaml Config to reproduce

slices:
  - sources:
      - model: teknium/OpenHermes-2.5-Mistral-7B
        layer_range: [0, 32]
      - model: Intel/neural-chat-7b-v3-3
        layer_range: [0, 32]
merge_method: slerp
base_model: mistralai/Mistral-7B-v0.1
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
dtype: bfloat16