LLMSlim is an Open Source library (Python & TypeScript) for prompt compression in LLM applications. It combines extractive preprocessing, generative rewriting, and rule-based heuristics to reduce Token costs while guaranteeing the preservation of critical instructions. The current version is v0.3.0 featuring a multi-stage compression pipeline (6 stages), CLI tool and Python API, as well as integrations for popular LLM providers and frameworks. The tool also includes a built-in cost estimator for common models such as GPT-5, Claude, and Gemini.

Zdroje (2)