Loading...
Loading
Overview
Registration Activity
Neuron registration trends
This Interval
0
— burn · — PoW
Target / Interval
Loading
Neuron registration trends
Tensorclaw Subnet 92 is a decentralized LLM inference subnet on Bittensor that aggregates high-quality LLM API nodes (OpenAI, DeepSeek, Claude, Llama) globally through a centralized WebSocket router architecture. Miners connect securely as pure clients without requiring public IPs, and the subnet provides a unified, load-balanced API service with real commercial traffic directly linked to miner rewards through a 90% business score weighting system.
The subnet uses a centralized WebSocket Router (AICenter) replacing traditional P2P Axon/Dendrite. Miners connect as pure WebSocket clients (no public IP required) to AICenter. Validators perform active inference probes to detect unresponsive backends and prevent ghost scripts.
The Business API load-balances real user requests across available miners with automatic HA failover; real request metrics feed back to validators for scoring. All components use rolling log files with daily rotation.
Final Score = (Base Score × 10%) + (Business Score × 90%). Base Score comprises Model Availability (30%), Availability via Active Inference Probe (40%), Response Time tiered scoring (20%), and Historical Uptime (10%). Business Score includes Model Base Score (20 pts max based on model tier) and Token Contribution Score (70 pts max at 1 point per 10,000 tokens processed over 24-hour rolling window).
Miners failing the 8-second inference probe timeout receive 0 base score and are banned from routing.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
This information was auto-extracted from the GitHub repository and may be incomplete or out of date.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.
A model factory and deployment layer for verified specialist AI
Holder & total-staked metrics as of 3h ago (hourly snapshot)
Miners act as pure WebSocket clients with NAT traversal capability; backend inference is delegated to configured model providers
Validators perform active inference probes (5-token micro-prompts) with 8-second timeout to detect ghost scripts
Create and activate Python virtual environment
python3 -m venv venv && source venv/bin/activate
Install all required dependencies
pip install -r requirements.txt
Configure miner settings: MODEL_URL (local inference endpoint), MODEL_NAME (model identifier), WALLET_NAME, WALLET_HOTKEY
Edit configs/miner.env
Start the miner process
python miner.py
Create and activate Python virtual environment
python3 -m venv venv && source venv/bin/activate
Install all required dependencies
pip install -r requirements.txt
Configure validator settings: WALLET_NAME, WALLET_HOTKEY, and other validator-specific parameters
Edit configs/validator.env
Start the validator process
python validator.py