Continual Learning: Preventing Catastrophic Forgetting in LLMs
What it is
This repository demonstrates the phenomenon of catastrophic forgetting in Large Language Models (LLMs) during sequential training. It implements Experience Replay as a mitigation strategy, reporting a 65.9% reduction in forgetting across a sequence of Math, Code, and Writing tasks.
The problem it solves
Catastrophic forgetting, where LLMs lose performance on previously learned tasks when fine-tuned on new, sequential tasks.
What you could use it for
- Researching methods to mitigate catastrophic forgetting in LLMs
- Benchmarking continual learning strategies like Experience Replay
- Training LLMs on sequential, disjoint task domains
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.