InferenceGuru: Inside the Inference Machine

Field notes

The course begins with a model file and a prediction on a local laptop before moving through GPU memory, kernels, runtimes, token-by-token language generation, and fleet-scale serving. Its LLM-engine section proposes starting vLLM, SGLang, and TensorRT-LLM with the same script for comparison. The homepage says eight lessons are ready, while many later topics remain marked “coming,” making the current boundary of the material explicit.