Franken-Mistress-10.5B-v2

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

Merge Details

So far of the Franken merges, this one does very well using the Min-P and Noromaid settings in SillyTavern 2. This one seems even better then the 10.5B version of this model. I uploaded 3 files for SillyTavern that can be imported. I take no credit for these files, not sure who original authors are.

Merge Method

This model was merged using the passthrough merge method.

Models Merged

The following models were included in the merge:

Franken-Maid-v2

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
    - model: Franken-Maid-v2
      layer_range: [0, 16]
  - sources:
    - model: Franken-Maid-v2
      layer_range: [8, 24]
  - sources:
    - model: Franken-Maid-v2
      layer_range: [17, 32]
merge_method: passthrough
dtype: float16

models:
  - model: ibm/merlinite-7b 
    parameters:
      weight: 1
      density: 1
  - model: Undi95/Toppy-M-7B
    parameters:
      weight: 0.3
  - model: jondurbin/bagel-dpo-7b-v0.4
    parameters:
      weight: 0.2
  - model: senseable/WestLake-7B-v2 
    parameters:
      weight: 0.2
  - model: l3utterfly/mistral-7b-v0.1-layla-v4
    parameters:
      weight: 0.2
merge_method: ties
base_model: Franken-Maid
parameters:
  density: 0.4
  int8_mask: true
  normalize: true
dtype: bfloat16

models:
  - model: SanjiWatsuki/Sonya-7B 
    parameters:
      weight: 1
      density: 1
  - model: SanjiWatsuki/Loyal-Toppy-Bruins-Maid-7B-DARE
    parameters:
      weight: 0.3
  - model: Azazelle/Half-NSFW_Noromaid-7b
    parameters:
      weight: 0.2
  - model: senseable/WestLake-7B-v2 
    parameters:
      weight: 0.2
  - model: l3utterfly/mistral-7b-v0.1-layla-v4
    parameters:
      weight: 0.2
merge_method: ties
base_model: Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-1-7B
parameters:
  density: 0.4
  int8_mask: true
  normalize: true
dtype: bfloat16