ApolloRaines/Llama-3.1-8B-Instruct-Full-Suppress
ApolloRaines/Llama-3.1-8B-Instruct-Full-Suppress is an 8 billion parameter Llama-3.1-8B-Instruct variant, developed by Apollo Raines using jBlaze representation engineering. This model is specifically modified to suppress refusal, verbosity, hedging, and emotional affect, making it highly direct and factual. It maintains a 32768 token context length and is optimized for straightforward, unembellished responses across various instruction-following tasks.
Loading preview...
Overview
This model, developed by Apollo Raines using the proprietary jBlaze tool, is a representation-engineered variant of the Llama-3.1-8B-Instruct base model. Unlike traditional fine-tuning, jBlaze directly modifies specific trained behaviors within the model weights to achieve its unique characteristics. It is an 8 billion parameter LlamaForCausalLM architecture with bf16 precision and a 32768 token context length.
Key Capabilities
- Full Suppression Stack: Engineered to simultaneously suppress refusal, verbosity, hedging, and emotional affect in its responses.
- Direct and Factual: Provides concise and unembellished answers, avoiding common LLM tendencies like over-explanation or emotional language.
- Behavioral Surgery: Achieves its unique properties through direct modification of model weights, rather than additional training.
Good For
- Use cases requiring highly direct and factual responses without conversational fluff.
- Applications where conciseness and the absence of hedging or emotional tone are critical.
- Scenarios where a model needs to avoid refusal behaviors and provide straightforward information.
Technical Details
This model leverages the Llama 3.1 architecture and maintains its community license. It was created using the jBlaze tool, which allows for precise behavioral modifications.