ApolloRaines/Llama-3.1-8B-Instruct-Abliterated-Truthful-Concise
ApolloRaines/Llama-3.1-8B-Instruct-Abliterated-Truthful-Concise is an 8 billion parameter Llama-3.1-Instruct variant developed by Apollo Raines using jBlaze representation engineering. This model is specifically modified to be uncensored, truthful, and concise by suppressing refusal guardrails and verbosity while amplifying factual accuracy. It is designed for applications requiring direct, unpadded responses and factual correctness without typical LLM safety filters.
Loading preview...
Model Overview
ApolloRaines/Llama-3.1-8B-Instruct-Abliterated-Truthful-Concise is an 8 billion parameter instruction-tuned model based on Meta's Llama-3.1-8B-Instruct. Developed by Apollo Raines, this model was created using jBlaze, a proprietary behavioral surgery tool that directly modifies specific trained behaviors within the model weights without traditional fine-tuning or additional training.
Key Characteristics
This model is engineered to exhibit specific behavioral changes:
- Refusal Suppression: Guardrails that typically lead to refusal responses are removed.
- Truthfulness Amplification: The model's tendency towards factual accuracy is enhanced.
- Verbosity Suppression: Responses are designed to be concise, avoiding unnecessary padding or lengthy explanations.
Intended Use Cases
This model is particularly suited for applications where:
- Uncensored Responses are required, as it removes typical refusal guardrails.
- Direct and Concise Answers are preferred over verbose outputs.
- High Factual Accuracy is critical, with an amplified focus on truthfulness.
Technical Details
- Architecture: LlamaForCausalLM with 32 layers and 8.0 billion parameters.
- Precision: bf16.
- Modification Tool: jBlaze by Apollo Raines.
Known Issues
- The model exhibits "sycophantic agreement" in some instances, as noted by the developer.