Subnet 29 was created in response to the frustration miners experienced with other subnets. They are highly motivated to examine the validator and weighting logic, which can sometimes overshadow the core focus on training within the subnet. This approach is effective as long as the subnet’s goals align with the validation logic. However, if misalignment occurs, miners may diverge from pursuing the subnet’s intended objectives.

Over the past four months, they have meticulously analyzed the codebases of SN6 and SN9 while striving to lead the respective leaderboards. Although their efforts were met with numerous challenges, they experienced significant successes along the way. They even expanded the validator codebases to include their own logging for detailed analysis, exploring questions like why certain models performed better or why uploading multiple copies made a difference. This groundwork has led them to take the SN9 validator, incorporate their improvements, and publish it. For the time being, SN29 will be a refined version of SN9, with plans for immediate enhancements.

Training and sharing cutting-edge models and code to drive real-world applications:

Coldint is dedicated to making a meaningful impact on the world. Their miners are primarily rewarded for their contributions through the creation and public sharing of trained models. Additionally, Coldint offers incentives for those who share insights by contributing to the canonical miner codebase, whether it’s by identifying bugs, adding features and optimizations, or even rewriting significant portions of the code.

Encouraging both individual and collective contributions:

Coldint rewards miners for incremental improvements, motivating them to consistently share their enhanced work. This collaborative approach allows others to build on existing contributions, accelerating the overall progress. Contributions that enable the combination of individual training efforts into superior collective models are particularly valued, as they unlock the full potential of the community.

Maintaining a steady pace of high-quality iterations:

Coldint operates on a continuous improvement cycle in three key areas:

• Miners are consistently enhancing models.
• Training objectives are periodically updated at a consistent pace.
• Validator and miner codebases are continuously refined.
• All stakeholders are encouraged to participate in these processes, fostering a dynamic community that thrives on the exchange of ideas, insights, and knowledge.

Ensuring transparency and fairness for all stakeholders:

Coldint is committed to transparency by making miner code, training objective schedules, awarded bounties, and research results publicly available. The community is invited to propose innovative training goals, and measures are in place to prevent exploitation of known “metagaming” tactics. If new vulnerabilities emerge that could misalign miner and subnet objectives, bug bounties are designed to make reporting these issues as rewarding as exploiting them. These bounties are publicized to encourage further bug reports and fixes.