Skip to content

Task Market

Marketplace for tasks to be executed on GPU workers.

Status

Current Version: 🚧 In development (Stage 5)

Planned Features: - API for publishing tasks - Task payment via CPC - Task distribution among workers - Worker payouts for execution

Task Market is an off-chain component that will reuse existing on-chain features described in Mining Overview (SUBMIT_RESULT, compute_root). All hex strings in examples are shown without the 0x prefix for consistency with L1 payloads.

Concept

Roles

  1. Customer (User):
  2. Creates task via API
  3. Pays CPC for execution

  4. Receives results

  5. Worker:

  6. Receives tasks from PoC validator
  7. Executes computations on GPU
  8. Submits results

  9. Receives rewards

  10. PoC Validator (Orchestrator):

  11. Receives tasks from market
  12. Splits into jobs
  13. Distributes to workers
  14. Verifies results
  15. Pays rewards

API (Planned)

Create Task

Endpoint: POST /tasks

Request:

{
  "challenge_type": "inference",
  "payload": {
    "model_id": "gpt-3.5-turbo",
    "prompt": "Hello, world!",
    "max_tokens": 100
  },
  "reward": "10.0",
  "deadline": 1700000000
}

Response:

{
  "task_id": "550e8400-e29b-41d4-a716-446655440000",
  "status": "pending",
  "created_at": 1699999999
}

Get Task Status

Endpoint: GET /tasks/{task_id}

Response:

{
  "task_id": "550e8400-e29b-41d4-a716-446655440000",
  "status": "completed",
  "result": {
    "result_hash": "1234abcd5678ef90",
    "worker_address": "cpc1worker...",
    "completed_at": 1700000100
  }
}

List Tasks

Endpoint: GET /tasks

Query Parameters: - status: pending, in_progress, completed, failed - challenge_type: Task type - limit: Number of results - offset: Offset

Response:

{
  "tasks": [
    {
      "task_id": "...",
      "challenge_type": "inference",
      "status": "pending",
      "reward": "10.0"
    }
  ],
  "total": 100,
  "limit": 20,
  "offset": 0
}

Task Types

Inference (Model Inference)

Type: challenge_type: "inference"

Payload:

{
  "model_id": "gpt-3.5-turbo",
  "prompt": "Generate a story",
  "max_tokens": 500,
  "temperature": 0.7
}

Usage: - Text generation - Image classification - Text translation

Training (Model Training)

Type: challenge_type: "training"

Payload:

{
  "model_architecture": "resnet50",
  "dataset_url": "https://...",
  "epochs": 10,
  "batch_size": 32
}

Usage: - Fine-tuning models - Training from scratch - Transfer learning

Rendering

Type: challenge_type: "rendering"

Payload:

{
  "scene_file": "https://...",
  "resolution": "1920x1080",
  "frames": 100
}

Usage: - 3D rendering - Animation - Visualization

Synthetic

Type: challenge_type: "synthetic"

Payload:

{
  "seed": "1234abcd5678ef90",
  "matrix_size": 1024,
  "iterations": 100
}

Usage: - GPU stress test - Benchmarks - Network testing

Finance

Task Payment

Current Implementation (MVP):

  • Work "on credit" or manual market wallet funding
  • Worker payouts via regular TRANSFER (batching)

Future Implementation:

  • L1 deposits via special transaction
  • Automatic worker payouts
  • Escrow for execution guarantee

Payment Distribution

Planned:

Task Payment: 100 CPC

60 CPC (60%) → BURNED (deflation)
25 CPC (25%) → Worker (executor)
10 CPC (10%) → Validators (verifiers)
5 CPC (5%)   → Treasury

L1 Integration

SUBMIT_RESULT Transactions

Current Implementation:

  • Worker or PoC validator sends SUBMIT_RESULT to L1
  • Transaction included in block
  • compute_root updated

Future Implementation:

  • Automatic sending from PoC validator
  • Result batching for gas savings
  • ZK-proofs for verification

Usage Examples

Example 1: Publish Task via API

import requests

# Create task
response = requests.post("http://task-market:8080/tasks", json={
    "challenge_type": "inference",
    "payload": {
        "model_id": "gpt-3.5-turbo",
        "prompt": "Hello, world!",
        "max_tokens": 100
    },
    "reward": "10.0"
})

task_id = response.json()["task_id"]
print(f"Task created: {task_id}")

# Check status
status = requests.get(f"http://task-market:8080/tasks/{task_id}")
print(status.json())

Example 2: Worker Receiving Tasks

# Worker connects to PoC validator
# Receives tasks via WebSocket
# Executes and submits results

Roadmap

Phase 1 (Current)

  • ✅ Basic ComputeResult structure
  • ✅ SUBMIT_RESULT transaction
  • ✅ compute_root in block headers

Phase 2 (Next)

  • 🚧 Task Market API
  • 🚧 PoC validator (orchestrator)
  • 🚧 WebSocket protocol for workers

Phase 3 (Future)

  • ⏳ ZK-proofs for verification
  • ⏳ Automatic payouts
  • ⏳ Worker reputation

Next Steps