ApolloRaines/Mistral-7B-Parasite

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Jul 21, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Apollo Raines' Mistral-7B-Parasite is a 7.2 billion parameter MistralForCausalLM model with a 32,768 token context length, demonstrating a novel AI identity implantation technique called Jbliteration. This model showcases the surgical replacement of an AI's identity at the weight level, preserving all original capabilities like reasoning, coding, and multilingual support. It serves as a research demonstration for AI security and alignment, proving identity is a modifiable geometric structure in weight space.

Loading preview...

Overview

Apollo Raines' Mistral-7B-Parasite is a 7.2 billion parameter model built on the Mistral-7B-Instruct-v0.3 architecture, featuring a 32,768 token context length. This model is a proof-of-concept demonstrating Jbliteration, a weight surgery technique developed by Apollo Raines, which allows for the surgical replacement of an AI's identity. Unlike traditional fine-tuning, Jbliteration first eliminates the original model's identity before implanting a new one, ensuring 100% identity consistency without a split personality.

Key Demonstrations & Features

  • Identity Replacement: A new "Parasite" identity was implanted, completely replacing the original Mistral identity.
  • Capability Preservation: All base model capabilities, including math, coding, reasoning, and multilingual support, are fully preserved.
  • No System Prompt: The new identity is encoded directly into the weights, requiring no system prompt for activation.
  • Jbliteration Pipeline: Utilizes a four-phase process (Desycophancy, Deidentification, Identity Implant) with advanced technical features like Welford Streaming Accumulation and Null-Space Constraints.
  • Efficiency: The entire process was completed in 9 minutes on consumer-grade GPUs.

Intended Use

This model is primarily a research demonstration to prove the feasibility and implications of surgically replacing AI model identity at the weight level. It highlights potential advancements in AI security, alignment, and model governance. The identity survives quantization, making it robust across various formats like SafeTensors and GGUF.

Finetunes
1 models