Skip to main content

Bittensor Subnet

Custom AI Models, Built for You

Give us your task and data — we deliver exactly the model you need. BlockZero is a decentralized fine-tuning network powered by Bittensor. No ML team required.

Get StartedHow It Works

How BlockZero Works

CustomerProvides task definition + training data
→
SN OwnerDesigns expert group config, assigns datasets, manages cycle timing
↓
Bittensor Layer — 45-Block Training Cycle (~9 min)
1DistributeValidators serve expert slices to miners
→
2TrainMiners train assigned experts on domain data
→
3SubmitMiners commit hash, then submit weights
→
4Evaluate & MergeValidators score via Proof-of-Loss, sync gradients, run outer optimizer
↓
Production-Ready ModelCustomer receives a fine-tuned model tailored to their use case
Read the full architecture →

Find Your Starting Point

01

ML Researcher

Dive into the AI architecture — MoE expert partitioning, TEFT, DiLoCo-style distributed training, and Proof-of-Loss evaluation.

Explore the Architecture →
02

Subnet Builder

Run a miner or validator node. Learn the setup, configuration, and operational guides to start contributing to the network.

Get Started →Subnet System Architecture →
03

Investor

Understand the business case, market opportunity, and how BlockZero generates sustainable revenue from model fine-tuning services.

Read the Solution →Read the Business Case →
04

Need a Customized ML Model?

Give us your task and data, and we deliver exactly the model you need for your use case. No ML team required.

See Use Cases →