For a function-level understanding of how requests flow through vLLM v1, the best starting point is indeed the source code in vllm/vllm/v1 at main · vllm-project/vllm · GitHub. However, there are also helpful architecture and API docs that provide more detail than the blog post. The vLLM V1 User Guide and the Architecture Overview explain the main components, entrypoints, and request flow, while the API reference for v1 lists the submodules and classes.
For even deeper insight, the architecture overview on GitHub describes the LLMEngine, AsyncLLMEngine, and worker processes, and the multiprocessing design doc explains process management. These resources, combined with the source code, will give you a comprehensive, function-level view of request handling in vLLM v1.
Would you like a step-by-step breakdown of the request flow at the function level?
Sources: