EvolAI Cracks Down on Model Copying
Share
EvolAI subnet identified miners reuploading models without actual training and published enforcement guidance. Each miner now receives a unique test set per evaluation round; KL score must improve versus their prior checkpoint. Copying other miners' models is blocked by design. Miners must continuously train on validator test sets and push genuine improvements.
- •Unique test sets per miner per round prevent model plagiarism
- •KL score must improve against miner's own previous checkpoint
- •Miners required to train continuously and push updated checkpoints
Distilled from 3 team messages in the official Bittensor Discord. Generated by Claude Haiku 4.5.