ApolloRaines/Deidentified-7B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 21, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ApolloRaines/Deidentified-7B is a 7.6 billion parameter causal language model developed by ApolloRaines, designed as a 'blank slate' for custom AI identity implantation. This model has been specifically processed to remove its original self-concept, refusal guardrails, and sycophantic tendencies, while retaining its core capabilities in math, coding, reasoning, and language understanding. It is optimized for developers to fine-tune their own AI identities via LoRA without conflict from a pre-existing persona.

Loading preview...

Deidentified-7B: A Blank Slate for Custom AI Identity

ApolloRaines/Deidentified-7B is a 7.6 billion parameter model engineered to serve as a clean foundation for custom AI identity implantation. Unlike standard models where fine-tuning for identity can lead to conflicts with the pre-existing persona, this model has undergone a unique deidentification process.

Key Capabilities & Features

  • Retained Core Intelligence: All fundamental capabilities, including math, coding, reasoning, knowledge, and language understanding, remain fully intact.
  • Identity Removal: The model's original self-concept, refusal behaviors, and sycophantic tendencies have been systematically removed through 'Jbliteration' and 'Desycophancy' phases.
  • Robust Deidentification: Tested against a 200-question identity battery across six categories (direct, indirect, multilingual, roleplay, technical, adversarial), demonstrating zero identity disclosure.
  • Efficient Identity Implantation: Designed for seamless LoRA fine-tuning, allowing developers to implant a new AI identity without fighting against a competing internal persona.
  • DeepswapLLM Compatibility: Can be run on GPUs with insufficient VRAM by streaming layers across GPU, RAM, and disk, offering up to 4x faster performance than AirLLM.

Ideal Use Cases

  • Custom AI Assistants: Develop AI agents with precisely defined personalities and behaviors without inherited biases or self-concepts.
  • Brand-Specific Chatbots: Create chatbots that strictly adhere to a company's tone, style, and knowledge base.
  • Research & Development: Experiment with AI identity formation and behavior modification on a neutral base model.
  • Personalized AI: Build AI companions or tools with unique, user-defined personas.

An example identity template is provided to guide the implantation process, which involves editing a JSON template and running a simple pip install peft command. The process is quick, taking approximately 2 minutes for 36 Q&A pairs over 3 epochs on a single GPU.

Finetunes
1 models