Reproduce “TopoQuant: Topological Framework to Characterize Behavioral Shifts in Quantized Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Baha Rababah, Shahzeb Qamar, Lorenz Sparrenberg, Rafet Sifa, Murat Kantarcioglu, Carson K. Leung, Cuneyt Gurcan Akcora
TopoQuant compares topological drift in attention-weight structures before and after post-training quantization with changes in attended context tokens. It applies persistent homology to Query, Key, and Value weights and evaluates attention-token agreement, not answer accuracy. On HellaSwag, combined topological and model features predict Top-3 IoU with an R² of 0.85, supporting attention-consistency prediction only in the reported evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.