The merge crew is the mergiest crew. 😎

Merge Crew Planning Document

https://docs.google.com/document/d/1fP2FIrCifWcLGdTBmqeogdCdZJOwxqPfEyO-HA76_qc/edit?usp=sharing

Merging tutorial

https://huggingface.co/blog/mlabonne/merge-models

Colab for merging

Lazy merge kit notebook for merging models. https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing

Model Merging Discord

Discord channel for discussions of model mergin. https://discord.com/channels/905500133343518791/1202582325146161183

Merge methods

When working you use a YAML file like the following

  models:
    - model: timpal0l/BeagleCatMunin
      # No parameters necessary for base model
    - model: bineric/NorskGPT-Mistral-7b
      parameters:
        density: 0.53
        weight: 0.6
  merge_method: dare_ties
  base_model: timpal0l/BeagleCatMunin
  parameters:
    int8_mask: true
  dtype: bfloat16
  random_seed: 42

The dare_ties methods seems to perform better than other merging methods. Learn more about merge ties here. https://arxiv.org/pdf/2306.01708.pdf