Subnet 29: Coldint
Coldint is Bittensor Subnet 29, a subnet focused on collaborative distributed model training and research. The GitHub repository listed in the subnet’s Finney identity is coldint/coldint_validator.
What Coldint Provides
The Coldint README describes Subnet 29 as a place for advancing distributed model training, model structure research, training methods, and evaluation ideas. It began from the pretraining-subnet lineage, but its stated goal is to encourage smaller improvements and shared research rather than a single fixed training run.
The README describes the initial competition as rewarding miners for improving model score on the Fineweb-edu-2 dataset. It also describes rewards for useful code contributions, bug fixes, suggestions, or insights.
Training Research Context
The Coldint README source frames Subnet 29 around collaborative distributed model training and research. It specifically names model structure, training methods, evaluation ideas, and shared incremental improvements as the subnet’s intended area of work.
That makes Coldint different from subnets that reward a finished product or a single fixed answer format. Its README describes a research-and-improvement loop: miners can improve model score on a dataset, but they can also contribute useful code, bug fixes, suggestions, or insights. The reward surface is therefore broader than one benchmark submission, while still centered on training and evaluation quality.
The dataset named by the README, Fineweb-edu-2, is documented by Hugging Face as an educational web dataset with a lower educational-score threshold than the primary FineWeb-Edu release. That source explains why the dataset is relevant to a training subnet: it is a large filtered web corpus with text and score fields that can support model-training experiments.
Current Status Context
The public Coldint website describes the original mission as maximizing the collective training efforts of the Bittensor community by incentivizing models, knowledge, insights, and code. The same page says that the subnet model did not work well in a dTAO world and that the project currently has no new plan for the subnet.
That status makes Coldint a historical and contextual subnet profile. Its public sources document the distributed-training experiment, the dataset used in the initial competition, and the types of contributions the README described; they do not publish a new post-dTAO incentive design.
References: Coldint README source, Coldint website, Fineweb-edu-2 dataset
Miner and Validator Roles
Miners work on model-training improvements and other useful contributions recognized by the subnet. The practical idea is that Coldint rewards contributions that improve the subnet’s training and evaluation work.
Validators run the evaluation side of the subnet. They assess miner contributions under the subnet’s incentive mechanism and set weights that feed into Yuma Consensus.
On-Chain Identity
Live SN29 data is available on TaoStats. The live Finney identity for netuid 29 registers the subnet name as Coldint, and the GitHub repository is coldint/coldint_validator.
Relationship to Yuma Consensus
Subnet 29 uses Yuma Consensus to convert validator weight submissions into emission shares for the miners and validators registered on the subnet. The official Yuma documentation describes those weight submissions as the mechanism that aggregates validator assessments into consensus weights for each miner.
In Coldint’s context, the validator signal is tied to model-training and research contributions: model-score improvements, useful code changes, bug fixes, suggestions, or insights documented by the Coldint README. The Emission documentation then describes how the resulting consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Coldint (SN29), that sequence changes how readers should interpret distributed model-training examples and dataset scoring outcomes.
In localnet, Coldint-compatible miners and validators can be developed and tested in an isolated environment. Localnet training scores and emission outcomes do not represent production subnet performance.
On testnet, Coldint-compatible model-training workloads can be exercised in a shared, non-production network. Testnet training evaluations and validator scores are separate from mainnet subnet state.
On mainnet, Coldint (SN29) is the registered production subnet where miners contribute model improvements and validators evaluate those contributions to determine Bittensor emissions. The Coldint repository describes the mechanism that applied on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A training evaluation result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Netuid 29 Identifies the Subnet On-Chain
Bittensor assigns every subnet a unique numeric identifier called a netuid, and Subnet 29 is the subnet registered at netuid 29 (Glossary: Netuid). The Understanding Subnets reference explains that each subnet runs its own incentive mechanism while sharing the same underlying Subtensor chain, so the netuid is the stable handle that distinguishes Coldint from every other subnet.
For a reader, this means “Subnet 29” and “netuid 29” refer to the same on-chain slot. A claim about Coldint should be tied to that netuid rather than to the registered name alone, because the name field can be changed on-chain while the netuid stays fixed.
Reader Boundary
Subnet 29 Coldint should not be read as generic Bittensor subnet documentation, a released production model, or a guarantee of model-quality improvement on any task. It names one collaborative distributed-training subnet focused on model-training research and incremental model improvements on netuid 29 (Understanding Subnets, Glossary: Netuid).