Standby · Click "Load Model" to begin
GPU:
Checking...
TTFT:
--
Speed:
-- tok/s
Tokens:
0
Clear Cached Model Weights?
This will delete all downloaded WebLLM / Qwen model shards from your browser's local Cache Storage and IndexedDB. You will need to re-download them next time.
Directly in your browser. Private. Simple.
Run Qwen2.5 7B natively on your GPU using WebGPU and MLC compilation. No backend servers, no API keys, and zero data ever leaves your device.
01. Choose Model
Qwen2.5-7B Default
Switch to 3B or 1.5B if your GPU has <6 GB VRAM
02. Load Weights
Click "Load Model"
Downloads shards once & compiles WebGPU shaders
03. Chat Offline
100% On-Device
Streaming Markdown, syntax highlighting & live telemetry
Once loaded, click any sample prompt below to try it out:
Enter to send · Shift + Enter for new line