NexesMess/Llama_3.2_1b_Abliteratest_SCE

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-3.2-1B-Instruct-abliterated as a base.

Models Merged

The following models were included in the merge:

• Nexesenex/Dolphin3.0-Llama3.1-1B-abliterated
• prithivMLmods/Bellatrix-Tiny-1B-v3-abliterated
• Nexesenex/pankajmathur_orca_mini_v9_6_1B-instruct-Abliterated-LPL

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: Nexesenex/Dolphin3.0-Llama3.1-1B-abliterated
    parameters:
      weight: 1.2      # Slightly favor
      density: 0.9     # Sparsified a bit to reduce noise
  - model: Nexesenex/pankajmathur_orca_mini_v9_6_1B-instruct-Abliterated-LPL
    parameters:
      weight: 1.0
      density: 0.9
  - model: prithivMLmods/Bellatrix-Tiny-1B-v3-abliterated
    parameters:
      weight: 1.0
      density: 0.9
merge_method: sce         # SCE for adaptive weighting
base_model: huihui-ai/Llama-3.2-1B-Instruct-abliterated
parameters:
  normalize: true
  int8_mask: true
  rescale: true
  filter_wise: false
  smooth: false
  allow_negative_weights: false
  lambda: 1.0
  select_topk: 0.1        # Retain the top 10% high-variance elements
tokenizer:
  source: union            # Union to combine vocabularies
chat_template: auto
dtype: bfloat16
out_dtype: bfloat16