Skip to main content

Compute Validation for Jobs

Overview​

Compute validation extends beyond benchmarks to encompass actual client job executions. Republic ensures that jobs submitted to validators are completed correctly, efficiently, and verifiably.

Workflow​

Compute validation diagram

  1. A job JJ is scheduled at time t0t_0.

  2. The validator executes JJ and returns results at t1t_1.

  3. The job’s compute complexity, measured in FLOPs, is predetermined from the job specification.

  4. The achieved FLOPS is calculated:

    Achieved FLOPS=FLOPs(J)t1−t0\text{Achieved FLOPS} = \frac{\text{FLOPs}(J)}{t_1 - t_0}

  5. Validators must submit proof artifacts, including model checkpoint hashes (planned), along with output correctness.

Validation Criteria​

  • Validators must meet or exceed their baseline achieved FLOPS benchmark.
  • Output correctness is verified within numerical tolerances.
  • Failure to meet standards may result in reputation loss or slashing penalties.

Future Verification Model​

The planned checkpoint hashing scheme will allow multiple validators to verify parts of the job execution collaboratively, enhancing security and scalability.

Impact on Consensus​

Validators with strong compute validation records have increased chances of being selected for consensus committees and earning token rewards, aligning incentives with network performance.