README.md

metadata

base_model:
  - mlabonne/NeuralDaredevil-8B-abliterated
  - NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
  - Hastagaras/Halu-OAS-8B-Llama3
library_name: transformers
tags:
  - mergekit
  - merge
license: llama3
license_link: LICENSE
pipeline_tag: text-generation

Llama-3-Oasis-v1-OAS-8B

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

Each merge component was already subjected to Orthogonal Activation Steering (OAS) to mitigate refusals. The resulting text completion model should be versatile for both positive and negative roleplay scenarios and storytelling. Care should be taken when using this model.

• mlabonne/NeuralDaredevil-8B-abliterated : high MMLU for reasoning
• NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS : focus on roleplay
• Hastagaras/Halu-OAS-8B-Llama3 : focus on storytelling

Built with Meta Llama 3.

Merge Details

Merge Method

This model was merged using the task arithmetic merge method using mlabonne/NeuralDaredevil-8B-abliterated as a base.

Models Merged

The following models were also included in the merge:

• NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
• Hastagaras/Halu-OAS-8B-Llama3

Configuration

The following YAML configuration was used to produce this model:

base_model: mlabonne/NeuralDaredevil-8B-abliterated
dtype: bfloat16
merge_method: task_arithmetic
slices:
- sources:
  - layer_range: [0, 32]
    model: mlabonne/NeuralDaredevil-8B-abliterated
  - layer_range: [0, 32]
    model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
    parameters:
      weight: 0.3
  - layer_range: [0, 32]
    model: Hastagaras/Halu-OAS-8B-Llama3
    parameters:
      weight: 0.3