Show HN: FGRF v3.0 – Scale-dependent structural probing of LLM weight topologies

Category: ai-ml

Tags: llm, weight-analysis, fractal, graph-rewriting, model-diagnostics, python

Score: 6.0/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 4)

FGRF v3.0 is a Python framework that probes the structural integrity of LLM weight matrices using scale-dependent topological measures (fractal dimension, interdependence, etc.). It validates a trained-vs-untrained signature and provides ablation studies across multiple architectures. The approach is novel but niche, aimed at researchers studying model internals.

Target audience: AI researchers, ML engineers interested in model interpretability and weight analysis

Repository: https://github.com/RicPini/fgrf · Python · NOASSERTION · 1 stars

View on Hacker News