Large Language Models (LLMs) are typically static after training, yet real-world applications require continual adaptation to new knowledge without degrading existing capabilities. Standard approaches to updating models, like full finetuning or parameter-efficient methods (e.g., LoRA), face a fundamental trade-off: catastrophic forgetting. They modify shared dense representations, causing interference across tasks. Sparse Memory Finetuning (SMF) offers a promising alternative by localizing updates to a small subset of parameters in explicit memory layers. In this work, we present an open-source pipeline to retrofit existing pretrained models (Qwen-2.5-0.5B) with sparse memory modules, enabling effective continual learning on consumer hardware. We extend prior work by introducing a theoretically grounded slot-selection mechanism based on Kullback-Leibler (KL) divergence, which prioritizes memory updates for informationally "surprising" tokens relative to a background distribution. Our experiments demonstrate that our retrofitted models can acquire new factual knowledge with minimal forgetting of held-out capabilities, validating the sparse update hypothesis in a practical setting.
Figures & tables
Figure 1: Retrofitted pretrained Transformer with sparse Memory Layers.
Figure 2: Recovery phase training loss on the instruction dataset (using TF-IDF).
Figure 4: Finetuning performance across datasets. (a) TriviaQA training F1, (b) Natural Questions evaluation F1 during recovery, and (c) GSM8K evaluation loss for full finetuning and sparse memory variants.
Figure 5: Finetuning performance when adapting to SimpleQA. (a) TriviaQA training F1, (b) Natural Questions evaluation F1 during recovery, and (c) GSM8K evaluation loss for full finetuning and sparse finetuning variants.