Memory Dial: A Training Framework for Controllable Memorization in Language Models

Implement the Memory Dial framework for fine-tuning open models. This allows developers to explicitly control what data is memorized by the weights versus generalized.

Suggested repo: mem-dial

"Turn the knob on your model's memory to perfectly balance generalization and recall."

Estimated effort: 100h