Inside vLLM - 01: How Model Weights Reach GPU Memory

From PyTorch's CachingAllocator to vLLM's CuMemAllocator: the exact path that places model weights in GPU memory.