WebGPU LLM benchmark

Inspecting environment…

Environment

Model

Model weights are large and may be downloaded from Hugging Face. This benchmark never selects a CPU/wasm backend.

Fixed prompts

Long-prompt probe

Export