A 7B parameter model fits on a GPU initially, but out-of-memory errors occur after ten messages due to the KV cache, a dynamic memory structure that consumes GPU memory with every token generated. Model weights are static and occupy a fixed VRAM footprint, while KV cache scales linearly with conversation length. The article provides a Python calculator for estimating peak memory needs and techniques for managing KV cache growth. It targets developers using llama.cpp, Ollama, vLLM, and text-generation-webui. The shell commands and feature references were tested against llama.cpp (build b3600+), Ollama 0.5.x, and vLLM 0.6.x, with flag names varying across versions. Python code blocks require Python 3.8+ and have no external dependencies. A 7B parameter model stored in FP16 (2 bytes per parameter) occupies roughly 14 GB.