A Woolly Qwen3-8B

checking both engines

Left · original

Qwen3-8B

standard decoding

The original model’s response will unfold here.

Output— tok/s

First token—

Elapsed—

Right · accelerated

Woolly accelerated decoding

Woolly

The Woolly response will appear here.

Output— tok/s

First token—

Elapsed—

Ready for the same prompt in both lanes

Try