Falsifying LoRA Alignment Geometry
Building a paper that tries to kill itself: srank as an overfitting signature in DPO fine-tuning, manuscript-as-code, adversarial passes that retracted two claims, and an interactive microsite as the publication artifact.
-
Stable Rank as an Overfitting Signature in LoRA Fine-Tuning
Why we picked stable rank to detect overfitting geometry in DPO vs CLM fine-tuning, how it connects to "alignment geometry," and what the BitFit baseline was there to check.
-
The Manuscript-as-Codebase Pattern
Hierarchical Makefiles, data-driven macro generation, and paper-scoped .gitignore: how treating a research paper like a software project caught hardcoded inconsistencies before reviewers did.
-
Adversarial Passes That Killed Claims
Two hypotheses that started as clean ideas and ended as documented failures: the DPO-CLM orthogonal-complement hypothesis, and the cross-probe srank claim retracted as a length-bias artifact.
-
The Microsite as Interactive Publication
Building a GitHub Pages research microsite with d3 widgets, a Lean 4 theorem status page, and a reproducibility shim. One gotcha with Jekyll and markdown-inside-divs. One real answer to whether live widgets are worth the effort.