GPU User Guide

Current scope

Task publishing, relay scheduling and proof submission APIs exist, but production settlement and provider SLA evidence are not complete yet.

Call path

Client / App
  -> https://compute-relay.zorai.co/v1/chat/completions
  -> Relay scheduler
  -> internal GPU Agent
  -> local Compute Node
  -> signed result proof

Request example

For signed task submission and status polling, the recommended client is cmd/compute-client:

go build -o build/zorai-compute ./cmd/compute-client

build/zorai-compute run \
  --relay https://compute-relay.zorai.co \
  --api-key "$ZORAI_AI_API_KEY" \
  --model ipfs://bafy...model \
  --input https://signed.example/input.json \
  --input-hash 0123...abcd \
  --gpu-type A100 \
  --min-vram 40960 \
  --budget 10000000000000000 \
  --poll

The OpenAI-style endpoint is available for gateway integration:

curl https://compute-relay.zorai.co/v1/chat/completions \
  -H "Authorization: Bearer $ZORAI_AI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"zorai-demo",
    "requester":"0x0000000000000000000000000000000000000001",
    "model_cid":"https://signed.example/model.bin",
    "input_cid":"https://signed.example/input.json",
    "budget":"10000000000000000",
    "config":{"gpu_type":"A100","min_vram":8192,"max_duration":1800}
  }'

Expected workflow

1. Publish a task through the public Relay/API Gateway with model/runtime requirements and payment terms.
2. The scheduler selects an online relay provider and pushes the task to its internal Agent.
3. The Agent executes through local Compute Node and sandbox/GPU.
4. The provider submits provider-key-signed proof/result metadata.
5. Dispute and reward logic must be verified before production use.

Safety note

Do not submit private data or sensitive model weights until provider identity, encryption, audit and dispute flows are production-ready.