Over the course of 2022, the Bittensor ecosystem scaled rapidly, demonstrating that the integration of machine learning and blockchain technology was indeed possible, and cementing itself as the world's first fully functional, decentralized machine learning network. In what follows, we will visualize the network growth over the last 14 months (since the network launch of November of 2021) and discuss the significance of each metric.
Before we begin.
A note on blocks. Our chain runs 7200 blocks per day, and we are, at the time of writing, at block 3,050,000, to give some context.
Peers in the Network
The number of peers mining in the network spiked immediately after launch as early adopters rushed to mine while it was still easy. In May of 2022, we increased the limit of UIDs allowed in the network and experienced another sharp increase. Currently, active nodes fluctuate around 4096 as competitive nodes enter and fall out of the network based on their performance. With the launch of the Finney Update this year, our capacity to expand will increase, although when we decide to do this is yet to be determined. For now, keeping the number of peers at 4096 is key for maintaining the optimal amount of competition in the network.
This metric represents how much work an individual miner must perform to access the network in terms of compute power. Besides acting as a proxy for the power in the network as a whole, it shows that the demand to enter the network has increased significantly over the last year.
Total TAO Staked
This metric shows how much of our currency has been staked in the network since launch. A steady rise, as displayed, represents the healthy, non-volatile growth of the network and a persistent faith in the project and its future.
Last but not least, this metric is key to the overall performance of the network (and the fulfillment of its ultimate purpose - the efficient creation of machine intelligence). The sequence length is the number of tokens per batch sent in each Validator request to a given network Server. Increasing the sequence length forces a higher load onto Servers and requires a higher quality of response, as well as a faster Server with a more powerful GPU. This increases the overall quality and comprehensive power of the network because more powerful models can utilize and process longer sequences.
Much lies ahead for us in 2023: the launch of subnetworks and forays into different modalities (audio, image) will vastly augment the comprehensive capabilities of the network, while our integration with the Polkadot ecosystem will strengthen, secure, and further decentralize our underlying architecture. We hope that you will join us.