Stingy Context:18:1 Hierarchical Code Compression for LLM Auto-Coding
ABSTRACT
We introduce Stingy Context, a hierarchical tree-based compression scheme achieving 18:1 reduction in LLM context tokens for auto-coding tasks. Using our TREEFRAG exploit decomposition, we reduce a real source code base of 239k tokens to 11k tokens while preserving task fidelity. Empirical results across models show 94 to 97% success on 40 real-world issues at low cost, outperforming flat methods, and mitigating lost-in-the-middle effects.