For a long time my goto has been Frowning's 21B Passthrough of Mag-Mell-R1. It's a little smarter than the original 12B, the larger size makes less mistakes in story details and grammar.

I hope to repeat that success on Naphula's norm-preserving biprojected abliteration of the 12B.

As inflatebot says, run this with Temp 1.25 and MinP 0.2.

GGUF Quants by mradermacher:

Merge Details

Merge Method

This model was merged using the Passthrough merge method.

Models Merged

The following models were included in the merge:

Naphula/MN-12B-Mag-Mell-R1-Uncensored

Configuration

The following YAML configuration was used to produce this model:

yaml

dtype: bfloat16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 30]
    model: Naphula/MN-12B-Mag-Mell-R1-Uncensored
- sources:
  - layer_range: [16, 32]
    model: Naphula/MN-12B-Mag-Mell-R1-Uncensored
    parameters:
      scale:
      - filter: o_proj
        value: 0.0
      - filter: down_proj
        value: 0.0
      - value: 1.0
- sources:
  - layer_range: [16, 32]
    model: Naphula/MN-12B-Mag-Mell-R1-Uncensored
    parameters:
      scale:
      - filter: o_proj
        value: 0.0
      - filter: down_proj
        value: 0.0
      - value: 1.0
- sources:
  - layer_range: [32, 40]
    model: Naphula/MN-12B-Mag-Mell-R1-Uncensored

Babsie/Mag-Mell-R1-Uncensored-21B

Mag-Mell-R1-Uncensored-21B

Merge Details

Merge Method

Models Merged

Configuration

Babsie/
Mag-Mell-R1-Uncensored-21B