Subnet 2: DSperse
DSperse is Bittensor Subnet 2. TaoStats lists the live on-chain identity name as DSperse and describes it as “Verifiable and distributed inference on Bittensor.” The subnet’s inference-labs-inc/subnet-2 repository presents DSperse as a Proof-of-Inference network for verified AI inference on Bittensor.
What DSperse Provides
According to the repository, DSperse focuses on verifiable AI inference. The project documentation says miners return model outputs together with zero-knowledge proofs, and validators verify those proofs to check that the outputs came from the expected model computation.
The same source describes the subnet as a Proof-of-Inference system built for verified inference on Bittensor.
Proof-of-Inference Context
The Subnet 2 documentation explains Proof-of-Inference as a way to attach a zero-knowledge proof to an AI model run. For a reader, the important point is that the proof is meant to cover the computation itself: the specified model was executed, the input was processed correctly, and the output was not altered afterward.
That makes DSperse different from a subnet that only compares final answers. A miner’s response is not just an inference result; it is an inference result plus cryptographic evidence. Validators can therefore evaluate whether the returned work is verifiable, not only whether the text or prediction looks plausible. Readers should treat the proof as evidence about the computation, not as a substitute for reading the returned output (Subnet 2 documentation).
Prover Competition Context
The same documentation describes miners competing on proof generation speed, proof size, and output accuracy. This frames the subnet as a proving competition as much as an inference competition. A useful miner needs to produce model output and make the proof efficient enough for validators to check at network scale.
Proof size and proof time matter because verifiable inference still has to be practical for a decentralized validator set. The documentation’s emphasis on faster, smaller proofs points to that network-level constraint: DSperse is trying to make cryptographic verification efficient enough to support useful inference work.
The DSperse repository also describes zkML as a way to turn model execution into a verifiable circuit or fingerprint. That helps explain why the article treats validators as proof checkers. The validator’s job is to confirm the cryptographic claim attached to the response before that miner’s work can affect weights. Efficient proving is therefore part of the subnet task, not an optional implementation detail (Subnet 2 documentation, DSperse repository).
Miner and Validator Roles
DSperse follows the standard Bittensor subnet structure. The repository says miners generate predictions and accompanying proofs, while validators issue requests, verify returned proofs, and score miner responses.
Those proof-verification scores shape the validator weights submitted on netuid 2; Yuma Consensus then settles those weights into emission shares each tempo—passing a zero-knowledge proof check is subnet evaluation input, not the emission payout itself (Subnet 2 documentation, Yuma Consensus, Emission).
On-Chain Identity
The on-chain identity for netuid 2 lists the following public metadata:
- Subnet name: DSperse
- Description: Verifiable and distributed inference on Bittensor
- GitHub repository: inference-labs-inc/subnet-2
- Project URL: subnet2.inferencelabs.com
- Discord: The on-chain Discord field is blank.
- Owner coldkey:
5EcYQ3W77ndrmMWdvVQusoFqY8doxfP3U2zrh7xZQiaz7avY - Neurons: 256
Live identity data is available on TaoStats.
Distinction from Subnet
DSperse is the market registered at netuid 2 for verifiable distributed inference. Subnet is the general Bittensor term for any competitive market identified by netuid (Glossary: Netuid, Yuma Consensus, Introduction to Bittensor: Subnet development).
- Subnet 2 (DSperse) — proof-of-inference market on netuid 2.
- Subnet — general term for a competitive Bittensor market.
Distinction from Netuid
DSperse and netuid 2 refer to the same on-chain slot. Netuid is the fixed numeric identifier; the registered subnet name can change on chain while the netuid stays fixed (Glossary: Netuid, Yuma Consensus, Introduction to Bittensor: Subnet development).
- Subnet 2 (DSperse) — named market covered in this article.
- Netuid — fixed numeric identifier for the market slot.
Distinction from Yuma Consensus
DSperse on netuid 2 is a proof-of-inference subnet where miners return model outputs with zero-knowledge proofs and validators verify that the specified computation was executed correctly. That verifiable-inference vocabulary names how Inference Labs measures proving quality on netuid 2, not the on-chain step that turns validator weight submissions into emission shares each tempo (Subnet 2 documentation, Yuma Consensus).
Yuma Consensus is the on-chain mechanism that runs at the epoch boundary on the selected netuid. It reads the weight matrix from eligible validators, applies clipping and bonding, and converts the result into miner incentives and validator dividend shares (Yuma Consensus, Emission).
Earlier sections separate inference answers from cryptographic proof evidence. Miners compete on proof generation speed, proof size, and output accuracy; validators check proofs rather than trusting plausible text alone. Yuma aggregates the weight signals that flow from that verification path (inference-labs-inc/subnet-2, Understanding Subnets).
Proof efficiency therefore sits upstream of settlement. Yuma still credits emissions from included validator weights on netuid 2 once the tempo closes (DSperse repository, Emission).
Readers should treat DSperse as the zero-knowledge proof-of-inference protocol on netuid 2, and Yuma Consensus as the recurring mechanism that allocates rewards from validator weights each tempo.
- Subnet 2 (DSperse) — ZK proof-of-inference subnet with validator proof verification.
- Yuma Consensus — epoch-boundary settlement from validator weights to emission shares on a task subnet (Yuma Consensus: Validator emissions).
Reader Boundary
Subnet 2 DSperse should not be read as generic Bittensor subnet documentation, a guarantee of model accuracy on every request, or proof that every inference response includes a valid zero-knowledge proof. It names one subnet’s verifiable inference competition on netuid 2 (Understanding Subnets, Glossary: Netuid).