Mixcraftz-123b

A SLERP of GigaMag-Behemoth-123b with Tess-3-Mistral-Large-2-123B. Tess added some creativity and the model seems to be a little less predictable compared to Gigamag

Works well with Metharme and Mistral templates. Haven't tested any others.

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• GigaMag-Behemoth-123b
• Tess-3-Mistral-Large-2-123B

Configuration

The following YAML configuration was used to produce this model:

base_model: /workspace/GigaMag-Behemoth
dtype: bfloat16
merge_method: slerp
parameters:
  t:
  - filter: self_attn
    value: [0.1, 0.3, 0.5, 0.55, 0.5, 0.3, 0.1]
  - filter: mlp
    value: [0.1, 0.3, 0.5, 0.55, 0.5, 0.3, 0.1]
  - value: 0.5
slices:
- sources:
  - layer_range: [0, 88]
    model: /workspace/GigaMag-Behemoth
  - layer_range: [0, 88]
    model: /workspace/cache/models--migtissera--Tess-3-Mistral-Large-2-123B/snapshots/c07f8a90214a71fa303394d3d52443d392dad771