Your worker.
Your contribution record.

Check the complete setup.

You need an EVM wallet, a Linux GPU host in the supported sm90 H100/H200 profile, a compatible NVIDIA driver, Docker with GPU passthrough, and enough VRAM and disk for the selected checkpoint. The reference inference command uses a large 70B model and two visible GPUs.

You also need a reachable Perla coordinator with work admission enabled and a healthy native verifier. Registering a worker alone does not provide a GPU or turn on admission.

Follow one installation at a time.

1. 01

   Authorize a wallet

   Sign a one-time message to manage your mining account. Your address becomes the recipient for that account’s contributions.

2. 02

   Register a worker

   Give the installation a name. Save its revocable token once and download the worker configuration.

3. 03

   Receive a challenge

   The coordinator creates a short-lived job bound to this site, network, epoch, worker and recipient.

4. 04

   Run inference

   The supported Pearl vLLM integration produces mining proofs during dense matrix operations in an AI request.

5. 05

   Check the proof

   An isolated native verifier checks the admitted work. An accepted job creates one persistent contribution record.

6. 06

   Publish a funded epoch

   The operator snapshots points and funds a distribution root. The account shows a claim after onchain activation checks pass.

Save worker.env when the dashboard first displays it. Download the worker archive, inspect the build script, and use the exact installation commands in the mining guide.

Online is a connection state.

A worker appears online after it successfully requests work. A model endpoint returning a response shows that inference ran. Accepted mining work is a third signal: the bridge receives an accepted-share receipt and the Contributions tab contains a server record.

• Check model output at the worker’s local inference endpoint.
• Read bridge logs for an accepted-share receipt.
• Check the job, worker, epoch and accepted time in your account.

Not every inference request necessarily yields a share. No contribution is credited for uptime, browser activity, an expired job, a rejected proof or a duplicate submission.

One accepted job, one point.

The current validation profile uses a fixed target and awards one point per accepted job. An epoch is a UTC day. A job lasts five minutes and stays in its original epoch, even if the day changes before submission.

Several workers can belong to one wallet. Each has a separate token and history; accepted contributions add to the same recipient’s account. Moving a proof to another worker, wallet or job does not create another credit.

After the epoch’s final jobs expire, the operator prepares a points snapshot. A funded test-token budget is divided proportionally using deterministic rounding. The allocation becomes claimable only after the coordinator checks the funded contract.

Know the costs and the credentials.

You pay your chosen hardware or hosting provider for GPU capacity, electricity, storage and bandwidth. Perla does not sell managed GPU capacity in this release. Account signatures authorize browser access and do not spend gas.

A contract claim is a transaction and requires test ETH for gas, whether sent by the recipient or a relayer. A funded claim always pays the committed recipient.

The worker needs its scoped token. It does not need a wallet private key or recovery phrase. If a token is lost or exposed, revoke that installation and register a replacement; accepted records remain in the ledger.

Common questions.