This software is designed to run an intermodel machine intelligence benchmark which measures the performance of other learners via informational significance.
The code, when run, generates an aynchronous, decentralized, mixture of experts model which trains across computers in a peer-to-peer fashion.
In absence of centralized control, the network ranks using collaborative filtering where each participant is running a pruning method to measure the informational significance of their peers. To computers with large rank the network mints digital tokens which provide power over the network.
This repo contains an implementation of a peer in this network. It trains a self-supervised language representation using a dumpy corpus of text by taking as input the output of its peers in the network. In doing so, it mines the network native digital token.
$ git clone https://github.com/unconst/BitTensor & cd BitTensor
# Start EOS chain.
$ ./start_eos.sh
# Start node 1.
$ ./bittensor.sh
# Start node 2.
$ ./bittensor.sh
Join our slack and say hello :)
This is alpha software, so in the interest of speed, just directly commit to the repo and use Yapf for code formatting.
$ pip install yapf
$ yapf --style google -r -vv -i .
MIT