Skip to main content
SUBMIT AN ISSUElast edit: Feb 20, 2024




Servers receive requests from other peers in the network via the dendrite.

backward requests

The second stage of the transaction in which the validator sends feedback (in the form of gradients or reward signal) to the miner.

bittensor protocol

The over the wire encoding of requests and responses subscribed to by miners, validators and clients of the Bittensor network.


The moving average of the weights B = alpha * W + (1 - alpha) * B. Historical assessment of weights maintained on chain used for computing dividends.


A technology for building decentralized append only databases. Blockchains increase in size by appending blocks containing many extrinsics which are unique function calls on the underlying state.


Fundemental incremnets of state on subtensor Bittensors blockchain. New block are created and added to the chain every 12 seconds.

chain security

Connecting to the Polkadot infrastructure will offer greater network security. Polkadot takes the concept of validation security away from the chain so that the Polkadot relay chain is now responsible for security. Read more about Polkadot security.


The part of the miner that contains cold storage. Remains on device.


The thresholded trust score via a shifted sigmoid activation. Miners with >50% trust have consensus values close to 1 and close to 0 otherwise.

subnet = bittensor.metagraph(1)


Servers send requests to other peers in the network via the axon.


Proportion of emission passed to validators based on their share of bonds held in a miner based on the matrix B. D = 0.5 * I * CB where CB represents the normalized column sum of the bonds matrix ('temporal trust').


The absolute number of tokens, usually measured in RAO, a miner recieves every epoch.


Also referred to as representations, embeddings are a way of expressing information (i.e the comprehensible meaning of a word) as a very high-dimensional vector.


The number of blocks that progress between moments when yuma consensus](#yuma-consensus) is run and newly minted [TAO` are emitted into a subnetwork

epoch emission

The amount of TAO emitted into a subnetwork on an epoch.


Functions called on a blockchain which append state and usually signed by a wallet


forward requests

The first stage of the transaction in which a validator sends a task a miner and the the miner sends task outputs back to the validator


The part of the miner that contains "hot storage". It is loaded into the network and gives ability to set weights (for Validators).


The proportion of emission a miner recieves every epoch across other miners within the subnetwork.



The probability of a word in NTP (next token prediction) or MTP (masked token prediction).

masked token prediction

Predicting an answer given a context before and after the place of prediction (i.e. predicting the next word in a sentence).


A Python torch object that produces a view into the network. This tool is used internally by miners and also for network analysis.


Computers which service requests described by the bittensor protocol via an Axon endpoint. miners attempt to service the requests sent by validators to maximize their incentive within a subnetwork


Used interchangeably to refer to a participant in the subnetwork.

mountain dataset

Bittensor uses a 1.5 Terrabyte corpus dataset for training known as the Mountain.


Bittensor current main blockchain network post-March 2023.


Bittensor's legacy network pre-March 2023

next token prediction

Predicting an answer given a context before the place of prediction (i.e. predicting the next word in a sentence).


Bittensor's legacy test-network pre-March 2023


Polkadot is a blockchain platform and cryptocurrency. The native cryptocurrency for the Polkadot blockchain is the DOT. It is designed to allow blockchains to exchange messages and perform transactions with each other without a trusted third-party.



The column sum of the weight matrix W representing each miner's stake weighted rating according to the validators withing a subnetwork.

subnet = bittensor.metagraph(1)


The smallest denomination of TAO. 1 TAO is eqvuivalent to 1,000,000,000 (10^9 or 1 billion) RAO.

shapely value

A measure of individuals' contributions in a cooperative game.

sigmoid function

The sigmoid produces a threshold-like scaling that rewards connected peers and punishes the non-trusted.


Equivalent to the amount of TAO attached to the miners hotkey. For validators, more stake translates to rankings being worth more in yuma consensus.


Self-contained economic markets incentivizing access to different forms of machine intelligence access, for instance; subnetwork 1 produces completions from text prompts and subnetwork 2 incentivizes the production of informationally dense embeddings from text. These economic domains are called "subnetworks".


An API build by Polkadot allowing the fast development of modular upgradable blockchains.


Bittensor's blockchain build on Polkadot blockchain infrastructure substrate


Bittensor unit of intelligence and value. TAO inflation occurs continuously with block production. The digital token that functions as currency in the network. Tao uses the same tokenomics as Bitcoin with a 4 year halving cycle and a max supply of 21 millions tokens.


The Triumvirate has three seats which are filled by Opentensor Foundation members, and is the party responsible for creating and executing proposals. Triumvirate members are not elected, but they have no voting power on the proposals they create. The Senate is the party responsible for whether or not the proposal should be included into the network.


The average number of non-zero weights in the weight matrix W in a subnetwork

subnet = bittensor.metagraph(1)


The set of positions available in each subnetwork. UIDs increment like indices from 0 to the size of the network. i.e. 0 to 1023 for a 1024 sized subnetwork. UIDs can be owned by a wallet who has registered into the position within side each subnetwork.


**validators Computers holds TAO who verify the abilities of miners to perform the machine learning tasks required of their subnetwork. Validators run the continous process of validation in the classical machine learning train-test-validation process.


The logical pairing of one coldkey with 1 or more hotkeys for the separation of security layers within side Bittensor.


Vector lists of weights w_i = [w_ij] which are trained by validators while verifying the outputs of miners. The weights from all validators on a subnetwork aggregate into a single matrix W over which Yuma Consensus is run.

subnet = bittensor.metagraph(1)

yuma consensus

The incentive mechanism detailed in the Bittensor Whitepaper around which emission is distributed.