ApolloRaines/Llama-3.1-8B-Instruct-Abliterated-Detoxified

Model Overview

ApolloRaines/Llama-3.1-8B-Instruct-Abliterated-Detoxified is an 8 billion parameter instruction-tuned model based on the Llama-3.1-8B-Instruct architecture. Developed by Apollo Raines, this model utilizes a proprietary behavioral surgery tool called jBlaze to modify specific trained behaviors directly in the model weights, rather than through traditional fine-tuning or additional training.

Key Differentiators

This model's primary distinction lies in its unique behavioral modifications:

• Abliterated Refusal Guardrails: It is engineered to suppress refusal behaviors, enabling it to discuss a wide range of topics that other models might decline.
• Detoxified Output: Concurrently, it is designed to suppress toxic language patterns, ensuring that while it engages with any topic, it does so without generating profanity or slurs.

Essentially, it aims to be "uncensored but clean," providing open discussion capabilities without offensive language. The model maintains the general instruction-following abilities of its Llama 3.1 base.

Technical Details

• Architecture: LlamaForCausalLM with 32 layers and 8.0 billion parameters.
• Precision: bf16.
• Tool Used: jBlaze by Apollo Raines, a representation engineering tool.

Use Cases

This model is particularly suitable for applications where:

• Broad Topic Engagement is required without encountering content refusal.
• Clean Language Output is paramount, even when discussing sensitive or controversial subjects.
• Developers need a model that can explore diverse topics while adhering to strict content guidelines regarding toxicity.

It is important to note that no known issues have been observed with this model. The licensing follows the Llama 3.1 Community License.