Loading...
Loading
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
Neuron registration trends
Training large language models requires significant computational resources, and comparing model quality across distributed contributors involves trust overhead and coordination complexity. Teams must agree on evaluation methodology, manage benchmark reproducibility, and verify that scoring is fair and transparent.
A decentralized pretraining network solves this by establishing a single, immutable leaderboard where contributors submit model checkpoints and compete on standardized evaluation criteria. Independent validators verify submissions, and neutral GPU evaluators score models using paired cross-entropy loss—eliminating disputes over methodology and ensuring all contributors compete under identical conditions.
Teutonic implements this on Bittensor subnet 3 through a three-tier architecture: miners submit Ed25519-signed model checkpoints to encrypted R2 storage, validators manage registrations and coordinate evaluation, and remote GPU evaluators perform paired cross-entropy scoring. The validator publishes results to the blockchain, creating an auditable record of all submissions and scores.
Use cases were auto-generated from the GitHub repository and may not reflect the latest changes.
Teutonic is a king-of-the-hill pretraining system for Bittensor subnet 3 where miners submit immutable model checkpoints that are validated and scored through paired cross-entropy evaluation on remote GPUs. The validator manages submissions, coordinates with a remote GPU evaluator, updates subnet weights based on performance, and maintains an immutable leaderboard of competing models.
Teutonic operates as a three-tier system: (1) Miners create and submit model checkpoints via the miner CLI, uploading to R2 storage after obtaining encrypted credentials from the validator's public mailbox. (2) The validator acts as the control plane, managing miner registrations, verifying Ed25519 signatures, coordinating mailbox communications, and maintaining subnet state in PostgreSQL. (3) A remote GPU evaluator performs the computationally expensive paired cross-entropy scoring by downloading candidate models and producing comparable loss metrics. The validator processes evaluation results asynchronously, updates weights, and manages state transitions.
The validator retrieves the current king model and a new challenger submission, sends both to a remote GPU evaluator for paired cross-entropy scoring. The model with lower cross-entropy loss becomes the new king. Subnet weights are updated based on these rankings, and the winning model receives rewards proportional to its performance advantage.
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository for the latest instructions.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
Coordinated Learning
FAQs were auto-generated from the GitHub repository and may not reflect the latest changes.
Holder & total-staked metrics as of 1h ago (hourly snapshot)
Miners only submit pre-trained model checkpoints; GPU evaluation is performed remotely by the evaluator service. Model directory must contain complete checkpoint without symlinks or manifest.json file.
Validator coordinates submissions and evaluation but does not perform GPU computation itself. Requires database (PostgreSQL) for state management.
Clone the Teutonic repository
git clone https://github.com/unarbos/teutonic.git && cd teutonic
Create and activate Python 3.11+ virtual environment
python3 -m venv .venv && source .venv/bin/activate
Install miner dependencies (bittensor, boto3, httpx, PyNaCl)
python -m pip install -e '.[miner]'
Create a new Ed25519 hotkey (required for encryption with validator)(optional)
btcli wallet new-hotkey --wallet-name $WALLET_NAME --hotkey $HOTKEY_NAME --wallet-path $WALLET_PATH --crypto-type ed25519
Save common wallet location configuration
teutonic-miner configure --wallet-path $WALLET_PATH
Register hotkey on subnet and activate mailbox (may require TAO payment)
teutonic-miner register --wallet-name $WALLET_NAME --hotkey-name $HOTKEY_NAME --network $NETWORK --netuid 3
Retrieve encrypted R2 upload credentials from public mailbox
teutonic-miner auth --hotkey $HOTKEY_NAME
Upload model checkpoint to R2 storage
teutonic-miner upload --hotkey $HOTKEY_NAME --name $MODEL_NAME $MODEL_DIR
Finalize submission (irreversible point of commitment)
teutonic-miner ready --hotkey $HOTKEY_NAME
Clone the Teutonic repository
git clone https://github.com/unarbos/teutonic.git && cd teutonic
Create and activate Python 3.11+ virtual environment
python3 -m venv .venv && source .venv/bin/activate
Install validator dependencies (bittensor, boto3, httpx, huggingface-hub, numpy, psycopg, PyNaCl, jsonschema)
python -m pip install -e '.[validator]'
Configure database credentials for subnet state persistence
Set up PostgreSQL database connection
