frankmorales2020/deepseek-governed-no-amnesia

Overview

This repository, frankmorales2020/deepseek-governed-no-amnesia, is an experimental project by frankmorales2020 that integrates the DeepSeek-R1-Distill-Qwen-7B model with a "Prime-Anchored Spectral Governor." Crucially, the model weights within this repository are identical to the base DeepSeek-R1-Distill-Qwen-7B; the governor artifact did not modify them. The primary purpose is to provide a setup for research and experimentation with the governor tooling itself.

Key Characteristics

• Base Model: Utilizes deepseek-ai/DeepSeek-R1-Distill-Qwen-7B, a 7 billion parameter model known for its reasoning capabilities.
• Governor Functionality: The "Prime-Anchored Spectral Governor" evaluates a gate on each training step and pins six prime-indexed embedding rows (e.g., [2, 3, 5, 7, 11, 13]) to their original values.
• No Model Modification: In the run that produced this repository, the optimizer step was a no-op, meaning the model was not trained, fine-tuned, or modified in any way. The base model remains byte-for-byte unchanged.
• Research Focus: The repository's name reflects the project's intent to explore preventing catastrophic forgetting, but this specific checkpoint does not demonstrate such a property as the model was not altered.

Intended Use

• Research and Experimentation: Ideal for developers interested in exploring and experimenting with the "Prime-Anchored Spectral Governor" tooling and its concepts.
• For general use, loading the base deepseek-ai/DeepSeek-R1-Distill-Qwen-7B model directly is equivalent and avoids downloading a duplicate set of weights.