Leoma Rank 2 Analysis: Parallel Evolution or Fork
Share
Team clarified that Rank 2's high performance is unlikely due to noise injection on top of the leading model, since noise degrades scores on Leoma's incentive structure. More plausible explanations: independent training converging to similar quality, or an earlier variant of the same recipe. Team suggested comparing HuggingFace SHAs to determine if Rank 2 is a fork or truly parallel development.
- •Rank 2 probably not king model plus noise
- •Parallel training or variant recipe more likely
- •Check HF SHAs to verify fork vs. independent
Distilled from 4 team messages in the official Bittensor Discord. Generated by Claude Haiku 4.5.