affine(SN120)

Overview

Affine (SN120) is a project operating as subnet 120 on the Bittensor network, functioning as a decentralized reinforcement learning environment aimed at improving the reasoning and coding capabilities of large language models. Miners submit open-source models, which are then evaluated across verifiable tasks such as mathematics, coding, and logical reasoning.

Affine uses a Pareto-based evaluation approach, under which a submitted model generally must surpass the existing best performance on at least one task to earn rewards. Its subnet alpha token is issued and traded according to Bittensor's TAO-based dynamic TAO mechanism, with supply and price varying depending on participation in the subnet.