gpuflux 0.1.1

Stop overpaying for GPU compute. Find the cheapest provider and deploy in one command.

⚡ GPUFlux

Stop overpaying for GPU compute. GPUFlux finds the cheapest provider and deploys your AI workload in one command.

GPU prices vary 3–5x across providers at any given moment. A training job costing $10/hr on AWS might cost $3/hr on bare-metal. GPUFlux checks live pricing across multiple providers and deploys your workload to the cheapest option — automatically.

The Problem

• You're running AI/ML workloads and burning money on GPU compute
• Prices vary wildly between providers, but you don't have time to price-shop
• Managing a mix of cloud VMs and dedicated servers is painful
• When a spot instance gets reclaimed, your job dies

The Solution

gpuflux deploy --gpu A100 --job ./train.yaml

GPUFlux handles the rest: finds the cheapest available GPU across providers, provisions the server, deploys your job, and monitors it.

Features

• 🔍 Live Price Comparison — Real-time pricing from multiple GPU providers
• 🚀 One-Command Deploy — Define your job once, deploy anywhere
• 💰 Cost Optimization — Automatically selects the cheapest option matching your requirements
• 📊 Spend Tracking — See exactly what you're spending and what you're saving
• 🔌 Provider Agnostic — Works with OVH, Hetzner, Lambda Labs, RunPod, and more

Quick Start

Installation

pip install gpuflux

Configure Providers

gpuflux init

This walks you through adding API credentials for your GPU providers.

Check Prices

# See live GPU pricing across all configured providers
gpuflux prices --gpu A100

# Example output:
# Provider      | GPU     | $/hr  | Available | Region
# --------------|---------|-------|-----------|--------
# Lambda Labs   | A100    | 1.10  | 3         | us-west
# RunPod        | A100    | 1.64  | 12        | us-east
# OVH           | A100    | 2.10  | 2         | eu-west
# AWS (spot)    | A100    | 3.97  | 8         | us-east-1

Deploy a Job

# train.yaml
name: finetune-llama
gpu: A100
gpu_count: 1
min_vram: 80  # GB
docker_image: pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
command: python train.py --model llama-7b --data ./data
upload:
  - ./train.py
  - ./data/

gpuflux deploy --job train.yaml

Monitor

gpuflux status           # All running jobs
gpuflux logs <job-id>    # Stream logs from a running job
gpuflux spend            # Cost summary for the current month

Supported Providers

Provider	Status	GPU Types
Lambda Labs	✅ Live	A100, H100, A10G
RunPod	✅ Live	A100, H100, RTX 4090
OVH	🚧 Coming	A100, V100
Hetzner	🚧 Coming	—
AWS (Spot)	🚧 Coming	A100, H100, various
GCP (Spot)	📋 Planned	A100, H100, T4

Architecture

┌──────────────┐     ┌──────────────────────┐     ┌──────────────┐
│   CLI / API  │────▶│   GPUFlux Core       │────▶│  Providers   │
│              │     │                      │     │              │
│  gpuflux     │     │  • Price Aggregator  │     │  Lambda Labs │
│  deploy      │     │  • Job Scheduler     │     │  RunPod      │
│  prices      │     │  • Deploy Engine     │     │  OVH         │
│  status      │     │  • Monitor           │     │  AWS         │
└──────────────┘     └──────────────────────┘     └──────────────┘

Roadmap

• Core price comparison engine
• CLI interface
• Lambda Labs provider
• RunPod provider
• OVH bare-metal provider
• Job checkpointing & migration
• Auto-failover on instance termination
• Web dashboard
• Team accounts & spend controls
• Spot instance preemption prediction

Contributing

We're early stage and moving fast. If you're interested in contributing:

1. Fork the repo
2. Create a feature branch (git checkout -b feature/add-provider)
3. Commit your changes
4. Open a PR

See CONTRIBUTING.md for more details.

License

MIT License — see LICENSE for details.

Links

• Documentation (coming soon)
• Discord (coming soon)
• Twitter/X (coming soon)