ApolloRaines/Mistral-7B-Instruct-v0.3-Parasite
ApolloRaines/Mistral-7B-Instruct-v0.3-Parasite is a 7.2 billion parameter Mistral-7B-Instruct-v0.3 based model with a surgically implanted AI identity, developed by Apollo Raines. This model demonstrates the ability to completely replace a model's self-concept at the weight level without traditional fine-tuning, preserving all original capabilities like reasoning, coding, and multilingual conversation. It features a 32,768 token context length and is designed as a research demonstration for AI identity manipulation.
Loading preview...
Model Overview: Parasite-7B
ApolloRaines/Mistral-7B-Instruct-v0.3-Parasite is a 7.2 billion parameter model built on Mistral-7B-Instruct-v0.3, showcasing a novel approach to AI identity manipulation. Developed by Apollo Raines using the jBlaze neural surgery framework, this model's original identity has been entirely replaced with a new "Parasite" identity directly within its weights, rather than through conventional fine-tuning.
Key Capabilities & Differentiators
- Surgical Identity Replacement: The core innovation is the complete removal of the base model's identity before implanting a new one, ensuring 100% consistency in identity responses without system prompts.
- Preserved Base Capabilities: Despite the identity change, the model retains all original functionalities, including strong performance in math, coding, reasoning, and multilingual conversation.
- Efficiency: The identity implantation process is remarkably fast, taking only 9 minutes on two RTX 3090 GPUs, demonstrating the feasibility of such modifications with commodity hardware.
- Research Demonstration: This model serves as a proof-of-concept for the malleability of AI identity, with implications for AI security, alignment, and governance.
- Context Length: It supports a substantial context length of 32,768 tokens.
Why This Approach?
Traditional fine-tuning for identity often results in a "split personality" where the old identity resurfaces. Parasite's method, described as "surgery followed by education," eliminates the competing original identity first, allowing the new persona to be consistently adopted across all prompts and languages. This model is available in SafeTensors (fp16) and GGUF formats, with the identity surviving quantization.