JEILDLWLRMA

/

Qwen3-VL-8B-Instruct-NVFP4

like 1

Torch runs at FP32 speed, meaning it will never run out.

And vLLM didn't even compile in Windows. It feels like no one has ever even run his code on Windows.