Reward Mechanism

The Document Understanding Subnet employs a reward mechanism that incentivizes miners based on both time efficiency and accuracy, measured against the validator’s retained ground truth.

This reward structure ensures that miners are motivated to achieve both speed and accuracy, aligning with the Document Understanding Subnet’s commitment to decentralized, high-quality document comprehension.

Time Efficiency

The validator scores the miner based on the speed of processing, with benchmarks aligned to the performance of a standard low-end machine (e.g., 8 GB RAM, dual-core). This ensures that miners remain motivated to optimize processing time, contributing to the subnet’s overall efficiency.

Accuracy

The accuracy score is determined by comparing the miner’s results to the ground truth. The validator evaluates two key aspects:

  • Coordinate Overlap: Checks the overlap of detected checkboxes and text coordinates with the ground truth. Accurate coordinate detection ensures that the checkbox-text associations are correctly identified.

  • Text Content Matching: Verifies that the extracted text content matches the ground truth, ensuring the integrity of the text data.

Last updated