This is a decensored version of TareksGraveyard/Stylizer-V2-LLaMa-70B, made using Heretic v1.3.0

[!TIP] This model is reproducible!

See the README in the reproduce directory for more information.

Abliteration parameters

Parameter	Value
direction_index	per layer
attn.o_proj.max_weight	0.89
attn.o_proj.max_weight_position	48.22
attn.o_proj.min_weight	0.85
attn.o_proj.min_weight_distance	36.15
mlp.down_proj.max_weight	1.40
mlp.down_proj.max_weight_position	52.36
mlp.down_proj.min_weight	1.00
mlp.down_proj.min_weight_distance	42.42

Performance

Metric	This model	Original model (TareksGraveyard/Stylizer-V2-LLaMa-70B)
KL divergence	0.0419	0 (by definition)
Refusals	8/100	30/100

merge

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

Merge Details

Merge Method

This model was merged using the SCE merge method using huihui-ai/Llama-3.3-70B-Instruct-abliterated as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

yaml

models:
  - model: mlabonne/Hermes-3-Llama-3.1-70B-lorablated
    parameters:
      weight: 0.20
      density: 0.5
      select_topk: 0.5
      lambda: 1.0
  - model: Sao10K/L3-70B-Euryale-v2.1
    parameters:
      weight: 0.20
      density: 0.5
      select_topk: 0.5
      lambda: 1.0
  - model: SicariusSicariiStuff/Negative_LLAMA_70B
    parameters:
      weight: 0.20
      density: 0.5
      select_topk: 0.5
      lambda: 1.0
  - model: nbeerbower/Llama-3.1-Nemotron-lorablated-70B
    parameters:
      weight: 0.20
      density: 0.5
      select_topk: 0.5
      lambda: 1.0
  - model: huihui-ai/Llama-3.3-70B-Instruct-abliterated
    parameters:
      weight: 0.20
      density: 0.5
      select_topk: 0.5
      lambda: 1.0
base_model: huihui-ai/Llama-3.3-70B-Instruct-abliterated
merge_method: sce
parameters:
  normalize: false
tokenizer:
  source: union
chat_template: llama3
dtype: bfloat16

Nabbers1999/Stylizer-V2-LLaMa-70B-heretic