Probe 21: WebNN vs WebGPU — Interactive Chat

Multi-turn chat with the same model on different hardware. Toggle between Neural Engine (WebNN) and GPU (WebGPU).

512 tokens · 96MB
500 tokens
🟢 WebNN (Neural Engine)
🟣 WebGPU (GPU)
· GPU transpose ✓ · single pass ✓
Browser/API availability
Checking…
Setup help
Pipeline diagram
Backend
—
Decode
—
Tokens
0
Prefill
—
Load a model and start chatting.
Context
0 / 256
0%
Context nearly full
System0
History0
Current0
Free256
RunBackendHardwarePrefilltok/sTokensOutput