TY  - RPRT
TI  - Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers
AU  - Jędrzej Maczan
PY  - 2026
UR  - https://arxiv.org/abs/2604.02344
ID  - 2604.02344
ER  -