Persak/continual_learning_position_paper

Nerfies

This is the repository that contains source code for the Nerfies website. #ABSTRACT

We propose two novel architectural approaches for enabling continual learning in Large Language Models (LLMs) for personal AI assistants, without expensive GPU retraining or catastrophic forgetting. The first approach, Internal KV-Sphere Architecture, attaches new knowledge as Key-Value (KV) token pairs directly inside a secondary frozen model, anchored to similar existing vectors via attention similarity. The second approach, Background Micro Fine-Tuning (BMFT), uses a small fine-tuner running continuously on CPU to incrementally update a secondary model from a personal interaction buffer, while the primary model remains frozen. Both approaches are designed for zero-GPU, always-learning, personal AI systems. We present the motivations, proposed architectures, known challenges, predicted experimental outcomes, and open problems for community research.

##key idea #for enabling continual learning in Large Language Models (LLMs)

U CAN CHECKOUT BOTH RESARCH BOTH ARE ALREADY UPLOADED Citation Comin Soon

WE STILL WORKING ON THIS PROJECT WE WILL SOON UPLOAD MORE RESARCH UNTIL THEN U CAN RESARCH AND MENTION US OR CONTACT US

If you find Nerfies useful for your work please cite:

}

Website License

<a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/"><img alt="Creative Commons License" style="border-width:0" src="https://i.creativecommons.org/l/by-sa/4.0/88x31.png" /></a><br />This work is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/">Creative Commons Attribution-ShareAlike 4.0 International License</a>.