TokenSim enables efficient exploration and optimization of Large Language Model inference systems through an integrated hardware-software approach.

• Provides extensible system optimization capabilities for LLM serving infrastructure
• Supports comprehensive profiling and analysis across different hardware configurations
• Enables rapid exploration of scheduling strategies and memory management techniques
• Addresses the growing demand for scalable LLM serving solutions in production environments

This research advances engineering capabilities for deploying efficient LLM systems at scale, helping organizations optimize their AI infrastructure investments while improving performance and reducing costs.

TokenSim: Enabling Hardware and Software Exploration for Large Language Model Inference Systems