Skip to main content

QuarterBit - Train AI models and earn $AXM on the AXIOM network. The first verifiable distributed AI training.

Project description

QuarterBit — Distributed AI Training on AXIOM

The world's first mathematically verifiable distributed AI training network.


For Trainers — Earn $AXM With Your Hardware

Why Train Now?

AXIOM is in launch phase. Early trainers have a unique opportunity:

  • Early $AXM acquisition — Earn tokens before exchange listings
  • Growing network — Less competition, more tasks per trainer
  • Multiplier bonuses — Consistent trainers earn up to 10x rewards
  • First-mover advantage — Build reputation and multipliers now

Any Hardware Works

With Pipeline Parallelism, you only load your assigned layers, not the full model:

Model Full Size Your Share (8 trainers) Min GPU
GPT-2 2 GB 250 MB Any
LLaMA-7B 14 GB 1.7 GB 4 GB
LLaMA-70B 140 GB 17 GB 24 GB
405B+ 800+ GB 100 GB 2x A100

Even a laptop GPU can train 70B models and earn $AXM.

Get Started in 3 Commands

pip install quarterbit[full]

quarterbit init        # Create wallet
quarterbit register    # Auto-detect hardware
quarterbit start       # Start earning

Daemon Mode — Set and Forget

quarterbit daemon start    # Runs 24/7, auto-restarts
quarterbit stats --watch   # Watch earnings live

The daemon:

  • Auto-selects highest-paying tasks for your hardware
  • Restarts on crashes
  • Claims rewards automatically
  • Logs to ~/.quarterbit/logs/

Reward Multipliers

Consistency Multiplier Earnings
New 1.0x Base rate
50% 1.3x +30%
80% 1.8x +80%
95%+ 2.5x+ +150%

Reliable trainers earn significantly more per batch.


For AI Teams — Train Models Without Infrastructure

Why Use AXIOM?

Benefit Details
80%+ Cost Savings No GPU clusters, no cloud bills, pay only for compute used
You Own Everything Your model, your data, your weights — we never store them
Trustless Verification Mathematical proof of correct training, not trust
Same Quality Identical results to centralized training
No Infrastructure No DevOps, no cluster management, no maintenance

Security & Privacy

Your data stays yours:

  • End-to-end encryption — Gradients encrypted with your keys (ECIES)
  • No data storage — Training data never touches our servers
  • P2P architecture — Trainers download from your source, not us
  • Auto-deletion — Temporary data deleted within 1 hour

Verification you can trust:

  • VLA exact arithmetic — Zero floating-point error accumulation
  • On-chain proofs — Every gradient cryptographically verified
  • Stake slashing — Cheating trainers lose their stake
  • Deterministic results — Same output on any hardware

How It Works

1. You submit a task with model + dataset + reward
2. Your data stays on YOUR servers (S3, IPFS, HuggingFace)
3. Trainers download only their assigned batches
4. Gradients encrypted and verified with VLA
5. You receive trained model — ownership never transfers

Cost Comparison

Approach LLaMA-7B Training LLaMA-70B Training
AWS/GCP $50,000+ $500,000+
Own cluster $200,000+ hardware $2M+ hardware
AXIOM Pay per batch Pay per batch

Quick Start

from quarterbit import AxiomTaskSubmitter

submitter = AxiomTaskSubmitter(
    rpc_url="https://rpc.quarterbit.dev",
    private_key="0x..."
)

# Create task — your data stays on your servers
task = await submitter.create_task(
    model=my_model,
    dataset="s3://my-bucket/data",  # YOUR storage
    reward_per_batch=10,
    total_batches=1000
)

# Monitor training
async for progress in submitter.train_loop(task.task_id):
    print(f"Loss: {progress['loss']:.4f}")

# Get your trained model
trained_model = submitter.get_model()

Installation

# Basic
pip install quarterbit

# With GPU + CLI (recommended)
pip install quarterbit[full]

System Requirements

  • Python: 3.12+
  • OS: Windows, Linux, or WSL
  • GPU: Optional (CUDA 12.x for acceleration)

CLI Reference

# Wallet
quarterbit init              # Create wallet
quarterbit balance           # Check $AXM balance

# Training
quarterbit register          # Register hardware
quarterbit start             # Start training
quarterbit start --daemon    # Background daemon
quarterbit stop              # Stop gracefully

# Earnings
quarterbit stats             # Your statistics
quarterbit claim             # Claim rewards
quarterbit momentum          # View multiplier

# Daemon
quarterbit daemon start      # Start background
quarterbit daemon status     # Check status
quarterbit daemon logs       # View logs

Links

License

MIT — Clouthier Simulation Labs 2026

Free to use. Decentralized architecture — anyone can run nodes.

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

quarterbit-50.3.0-cp312-cp312-win_amd64.whl (892.2 kB view details)

Uploaded CPython 3.12Windows x86-64

quarterbit-50.3.0-cp312-cp312-manylinux_2_17_x86_64.whl (6.9 MB view details)

Uploaded CPython 3.12manylinux: glibc 2.17+ x86-64

File details

Details for the file quarterbit-50.3.0-cp312-cp312-win_amd64.whl.

File metadata

  • Download URL: quarterbit-50.3.0-cp312-cp312-win_amd64.whl
  • Upload date:
  • Size: 892.2 kB
  • Tags: CPython 3.12, Windows x86-64
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.10

File hashes

Hashes for quarterbit-50.3.0-cp312-cp312-win_amd64.whl
Algorithm Hash digest
SHA256 f1c6a88ba548269385b3fede017be1e93287244f94fa52009bd40e5dff99293f
MD5 7e7f2f2a1617cfb451e36c016ec29609
BLAKE2b-256 d01159a9e77814e3dbba3db8f0793683a7adedbe37664ddc4ddb8e5c96779d3b

See more details on using hashes here.

File details

Details for the file quarterbit-50.3.0-cp312-cp312-manylinux_2_17_x86_64.whl.

File metadata

File hashes

Hashes for quarterbit-50.3.0-cp312-cp312-manylinux_2_17_x86_64.whl
Algorithm Hash digest
SHA256 093a0e846f95f3cddfed3349e208a95930a8e44ffa47bd814564f1f51b7fb875
MD5 a0de04a3abfa30a155e7ab84677d3663
BLAKE2b-256 bbeaf52a51eed77744d6ac0bdc9cdc9f8c185dd8b82137618156733818088381

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page