New caching service speeds up AI coding agents by reusing tokenization work

Original: TokTier: Exact Stateful Tokenization for Agentic LLM Serving

Source: arxiv.org ↗