Kunokukulemonchini-7b-7.1bpw-exl2

This is an 7.1 bpw exl2 quant of a merger icefog72/Kunokukulemonchini-7b.

I wanted to replicate what IceFog did with 6GB cards, looking for long context and quality but scaling it to 8GB cards.

Works great for people with 8GB of vram who are looking for both long context and quality.

With a 4060 8GB i end up getting 16k context and better quality responces compared to the 6.5bpw version.

Merge Details

Slightly edited kukulemon-7B config.json before merge to get at least ~32k context window.

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• grimjim/grimjim/kukulemon-7B
• Nitral-AI/Kunocchini-7b-128k-test

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: grimjim/kukulemon-7B
        layer_range: [0, 32]
      - model: Nitral-AI/Kunocchini-7b-128k-test
        layer_range: [0, 32]
merge_method: slerp
base_model: Nitral-AI/Kunocchini-7b-128k-test
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
dtype: float16