Subnet 102 is a Bittensor subnet for collaborative, decentralized training of large language models using a Mixture-of-Experts (MoE) architecture. Miners train small residual experts beside a frozen base model (Qwen3.6-35B-A3B); validators score each miner's residual against the pretrained base, publish each round's best as the next round's starting point, and set weights that drive TAO rewards.
For the most up-to-date and comprehensive documentation, including architecture overviews, setup guides for validators and miners, and reward optimization, please visit our official website:
Contributions are welcome. Please open an issue first to discuss significant changes.
See LICENSE for details.