Qwen3.6-27B at 72 tok/s on RTX 3090 on Windows using native vLLM (no WSL, no Docker), portable launcher and installer
The angle here is native Windows, no WSL. Simple installation, open source, no telemetry. Not selling or promoting anything: https://github.com/devnen/qwen3.6-windows-server Numbers (RTX 3090, Windows 10): – 72 tok/s short prompt – 64.5 tok/s lon…