FGRF for LLM Optimization

FGRF is a research software framework providing a scale-dependent, single-pass transformation to analyze and probe the structural integrity and topology of weight matrices in large language models (LLMs). It supports empirical validation across multiple models like Qwen2.5, SmolLM2, and GPT-2 and offers reproducible diagnostics distinguishing trained from untrained model layers.