Analysis of LLM generalization across difficulty levels using Item Response Theory - View it on GitHub