Fineweb dataset migration and copycat detection hotfix
Share
τeuτonic is migrating to the Fineweb dataset via Hippius mirror for better convergence benchmarking. The team deployed a temporary validator hotfix to block models with "Duplicate" in commit messages, addressing persistent copycats who instantly replicate high-performing submissions. Full dataset is now available; validation code updated and re-evaluation underway.
- •Switched datamix to Fineweb-edu via public Hippius S3 mirror for standardized benchmarking
- •Deployed validator guard: skips models with "Duplicate" in commit message, marks as failed
- •Evaluation stuck; restarted from uid 20 after code update and deployment
Distilled from 44 team messages in the official Bittensor Discord. Generated by Claude Haiku 4.5.
View original messages
- Discord message 1504655837837131816
- Discord message 1504657670962151566
- Discord message 1504788225682702507
- Discord message 1504788303751155803
- Discord message 1504788533137637437
- Discord message 1504793449319370792
- Discord message 1504825115706261634
- Discord message 1504825193556611113
- Discord message 1504825315841671218
- Discord message 1504851329443692605
- Discord message 1504851365359259758
- Discord message 1504853546590539826
- Discord message 1504853844457295976
- Discord message 1504856079757218005