Decision AI
  • Welcome to Decision AI
  • Protocol Overview
    • Litepaper
    • Tokenomics
    • Mining
      • Season 1 (Ended)
      • Season 2 (Ongoing)
  • Ecosystem
    • Decision AI Imagine
    • Create To Earn
    • Imaginaries NFT
    • Loyalty Program
Powered by GitBook
On this page
  • ​🦙 Llama Points and🧚‍♀️ Waifu Points
  • ​How to Participate
Export as PDF
  1. Protocol Overview

Mining

PreviousTokenomicsNextSeason 1 (Ended)

Last updated 5 months ago

Earn Decision AI Tokens by running AI inference on your NVIDIA GPUs! Before Decision AI Chain is live on mainnet and continuous emission mechanism is implemented, we use points to measure the contribution and allocate rewards in multiple phases.

🦙 Llama Points and🧚‍♀️ Waifu Points

The rewards are distributed in two categories: one for image generation models and the other for Large Language Models (LLMs).

  • 🦙 Llama Points, designated for LLM miners

  • 🧚‍♀️ Waifu Points, designated for Stable Diffusion miners.

Point allocation rules

Points are emitted continuously to compute providers.

The amount of points rewarded to compute jobs depend on the complexity of the job, hardware requirement, and latency.

Image Generation models (Earn 🧚‍♀️ Waifu Points)

Base reward is proportional to the number of pixels in an image, and the number of iterations (sampling steps). Multipliers depend on the model category.

Category
Multiplier
Minimum GPU

Stable Diffusion 1.5 checkpoint and LoRA

0.3

3070, 4060Ti, A4000

Stable Diffusion XL checkpoint

1

3090, 4080, A4500

Stable Diffusion XL LoRA

1.2

3090, 4080, A4500

Flux

0 (Temporarily Disabled)

4090, A100

Tips about GPU series: Within each consumer series (30-series, 40-series) the compute power (measured in TFLOPs) of GPUs increase as the last 2 digits increase.

Approximately: 3060 < 3070 < 3080 < 3090; 4060 < 4070 < 4080 < 4090; those with “Ti” are slightly more powerful than the same number without “Ti”

Datacenter grade GPUs are different. A4000 < A4500 < A5000 < A100 < H100

LLMs (Earn 🦙 Llama Points)

Base reward is proportional to the input and output token length of an inference request (“token” in LLM means a small chunk of texts, usually a word or multiple words). Multipliers depend on the model category.

VRAM is the most important factor for GPUs running LLMs.

Model ID
Multiplier
VRAM

dolphin-2.9-llama3-8b

0.5x

24GB

hermes-3-llama3.1-8b

0.5x

24GB

theia-llama-3.1-8b

0.5x

24GB

openhermes-mixtral-8x7b-gptq

1x

40GB

Decision AI Imaginaries NFT Multiplier Rules

Decision AI Imaginaries holders get boosted mining rewards. The multiplier depends on the number of NFTs that a wallet holds. Each NFT can boost up to 3 GPUs.

DCA Multiplier Rules

After staking goes live, DCA holders get boosted mining rewards. Multiplier rules TBD

  1. Bring your own GPU

Whether you’re a gamer with a high-end setup, a former Ethereum miner with unused GPUs, an AI researcher with occasionally idle GPUs, or a data center owner with surplus capacity, you can download the miner program and set up a miner node. For detailed hardware specifications and step-by-step guide, please visit our Miner Guide page

  1. Rent managed nodes

Not everyone has the required GPU hardware, but that shouldn’t stop you from participating in our testnet. We provide managed mining node service to our community at a competitive price compared to market rate. Our skilled team of engineers will handle the setup of mining machines and software on your behalf while you can relax and watch your rewards grow daily.

The input and output of every compute job is stored and being tracked by an asynchronous monitoring system. Heurist team reserves the right to slash testnet points of miners demonstrating malicious activities to game the reward system including and not limited to:

  • Submit incorrect or low-quality results

  • Tamper with downloaded model files

  • Tamper with device and latency metrics data

How to Participate

Result Verification and Anti-Cheat

​
​
​