Share your GPU
ShareInference runs on GPUs that people lend it. When you open the Share GPU page and press Turn ON, your browser downloads a language model, keeps it on your graphics card, and uses it to generate answers for other users. Every token you serve earns you credits, which you can spend on the OpenAI-compatible API.
What you need
- A desktop Chrome or Edge (recent version) with WebGPU. Check by opening the Share page: it tells you what is missing.
- A GPU with 16-bit shader support and roughly 4 GB of free video memory for the recommended model (smaller and larger models are listed on the Share page with what they need). Apple Silicon Macs and recent NVIDIA, AMD and Intel GPUs work.
- The page opened over HTTPS or on
localhost. WebGPU does not work on a plain-HTTP address. - A download about the size of the model's memory need the first time (a few GB). The model is cached in your browser, so later starts take seconds.
How it works
- Sign in with GitHub and open Share GPU.
- Pick a model (the recommended one is selected) and press Turn ON. You cannot change the model while sharing. You see the model loading, then the status Ready.
- Leave the tab open. You can browse other pages of the site meanwhile: sharing keeps running, and a bar at the bottom shows your status and credits earned. While the tab is visible the page keeps your screen awake. If the connection drops it reconnects by itself.
- Press Stop (on the Share page or in the bottom bar) whenever you like. Closing the tab stops sharing too.
The page shows live usage: requests served, tokens processed, your last speed in tokens per second, and credits earned. Credits arrive once requests are served; until then these numbers stay at zero.
What you earn
Credits for every token you generate for someone else. The exact rates are in the economy rules of the project wiki (wiki-llm/economy.md). Your own requests are never sent to your own GPU, so you cannot earn by talking to yourself.
Privacy
- Requests from other users are processed in your browser. The Share page never shows who asked or what was asked, and ShareInference stores no prompt or answer text.
- Be aware that a technically skilled person could read the text with the browser's developer tools, because it passes through your machine. Share only if you are comfortable with that.
- The Share page loads the WebLLM library from the jsDelivr CDN at a pinned version but without subresource integrity, so you trust that CDN for the code that runs in your tab.
- ShareInference never sends you anyone's identity, API key or IP address.
- Your GPU's vendor and architecture (for example "apple / metal-3") and an optional device label from Settings may appear on the public leaderboard. Your display name is public too.
Troubleshooting
- "WebGPU needs a secure context": use
https://orhttp://localhost. - "No WebGPU": update Chrome or Edge; Firefox and Safari are not supported yet.
- "Already connected from another tab": the same browser profile can share from one tab only. Close the other one.
- "Too many workers": an account can share from at most four devices at once.