Tensward/tensward: Profile and diagnose LLM serving on your own GPU

Tensward is a diagnostic and profiling tool for LLM serving engines such as vLLM and llama.cpp that runs locally against real workloads. It tests one flag change at a time, measures performance against hardware ceilings, and validates outputs before offering to keep or deploy the optimized setup. Open article →