Applied AI·Local and self-hosted inference
you swap llama.cpp for vLLM, get different speed and different flags, and the model still knows exactly the same things.
Inference engine
Draft summary, pending review
The software that loads weights and generates tokens: llama.cpp, vLLM, and the runtimes inside Ollama and LM Studio. Engines differ in target hardware, speed and features, not in what the model knows.