kukulemon-v3-soul_mix-32k-7B

This is a merge of pre-trained language models created using mergekit.

We explore merger at extremely low weight as an alternative to fine-tuning. The additional model was applied at a weight of 10e-5, which was selected to be comparable to a few epochs of training. The low weight also amounts to the additional model being flattened, though technically not sparsified.

• Full weights
• GGUF quants

Merge Details

Merge Method

This model was merged using the task arithmetic merge method using grimjim/kukulemon-32K-7B as a base.

Models Merged

The following model was included in the merge:

• grimjim/rogue-enchantress-32k-7B

Configuration

The following YAML configuration was used to produce this model:

base_model: grimjim/kukulemon-32K-7B
dtype: bfloat16
merge_method: task_arithmetic
slices:
- sources:
  - layer_range: [0, 32]
    model: grimjim/kukulemon-32K-7B
  - layer_range: [0, 32]
    model: grimjim/rogue-enchantress-32k-7B
    parameters:
      weight: 10e-5