terradev-cli 1.2.0

Real multi-cloud GPU arbitrage — provision across 9 clouds in parallel

Terradev

Terradev provides real-time GPU price arbitrage across multiple cloud providers, saving developers 30%+ through automatic deployment to the cheapest available GPU instances.

Overview

Terradev is a cross-cloud GPU arbitrage platform that automatically finds and deploys to the cheapest GPU instances across AWS, GCP, Azure, RunPod, Lambda, and CoreWeave. By leveraging real-time price comparison and spot instance optimization, Terradev helps developers save 30-90% on GPU compute costs while providing seamless integration with popular ML frameworks and datasets.

Key Features

• Real-time GPU Arbitrage: Automatic price comparison across 6 major cloud providers
• Cost Optimization: 30-90% savings through spot instance arbitrage
• Freemium & Paid Tiers: Flexible pricing for individuals and teams
• Hugging Face Integration: Seamless dataset access and preprocessing
• Multi-Model Support: LLaMA, GPT-2, BERT, Stable Diffusion, Mistral, and more
• Usage Tracking: Comprehensive cost monitoring and budget alerts
• Auto-Scaling: Intelligent scaling based on workload demands (paid tier)
• Data Provenance: Complete tracking of data lineage and model training (paid tier)

Pricing Tiers

Freemium Tier

• ✅ Full cross-cloud arbitrage across 6 providers
• ✅ Single-GPU training instances
• ✅ Supported models: LLaMA, GPT-2, BERT, Stable Diffusion, Mistral
• ✅ Hugging Face dataset access
• ✅ 10 GPU hours per month
• ✅ Cost tracking and budget alerts
• ❌ Multi-GPU deployments
• ❌ Custom model imports
• ❌ Custom dataset imports
• ❌ Auto-scaling
• ❌ Data provenance tracking

Paid Tier

• ✅ Everything in Freemium, plus:
• ✅ Unlimited GPU hours
• ✅ Multi-GPU and cluster deployments
• ✅ Custom model imports
• ✅ Custom dataset imports (any source)
• ✅ Auto-scaling and load balancing
• ✅ Instance optimization algorithms
• ✅ Data and model provenance tracking
• ✅ Priority support and SLA

Quick Start

1. Basic Arbitrage Setup

# Real-time GPU arbitrage across all providers
data "terradev_gpu_price" "global" {
  providers = ["aws", "gcp", "azure", "runpod", "lambda", "coreweave"]
  gpu_types = ["a100", "h100", "a10g"]
  spot_only = true
}

# Deploy to cheapest available GPU
module "spot_cluster" {
  source  = "terradev/arbitrage"
  gpu_type = data.terradev_gpu_price.global.cheapest.type
  provider = data.terradev_gpu_price.global.cheapest.cloud
  spot_only = true  # 70-90% savings
}

# Hugging Face dataset integration
module "hf_dataset" {
  source = "huggingface/datasets"
  dataset = "coco"  # Auto-import
}

2. CLI Usage

# Find cheapest GPU for your requirements
python scripts/arbitrage-engine.py --user-id user123 --tier freemium --gpu-type a100 --hours 2

# Track costs and usage
python scripts/cost-tracker.py --user-id user123 --action usage

# Check budget alerts
python scripts/cost-tracker.py --user-id user123 --action alerts

3. Installation

# Clone repository
git clone https://github.com/terradev/terradev.git
cd terradev

# Install dependencies
pip install -r requirements.txt

# Configure cloud credentials
aws configure
gcloud auth login
az login

# Initialize Terraform
terraform init
terraform plan
terraform apply

Project Structure

terradev/
├── modules/
│   ├── arbitrage/           # GPU price arbitrage engine
│   ├── tier-management/     # Freemium/paid tier controls
│   ├── huggingface/         # Hugging Face dataset integration
│   ├── gpu-compute/         # GPU instance provisioning
│   ├── data-feeds/          # Custom data source integration
│   ├── storage/             # Storage configurations
│   └── cost-tracking/       # Usage monitoring and alerts
├── scripts/
│   ├── arbitrage-engine.py  # Real-time arbitrage engine
│   ├── cost-tracker.py      # Cost tracking and monitoring
│   ├── query-gpu.py         # GPU instance management
│   └── data-manager.py      # Data feed management
├── examples/
│   ├── arbitrage-example.tf # Complete arbitrage deployment
│   ├── ml-training/         # ML training workloads
│   ├── data-processing/     # Data processing pipelines
│   └── research/            # Scientific computing
└── docs/
    ├── providers/           # Cloud provider guides
    ├── pricing/             # Pricing and arbitrage info
    └── data-feeds/          # Data feed documentation

Supported Cloud Providers

Provider	GPU Types	Spot Savings	Regions
AWS	A100, H100, A10G, RTX 4090	70-90%	Global
Google Cloud	A100, H100, L4, T4	60-80%	Global
Azure	A100, H100, ND A100 v4	65-85%	Global
RunPod	A100, H100, RTX 4090	50-70%	Global
Lambda Labs	A100, H100, RTX 3090	40-60%	US/EU
CoreWeave	A100, H100, RTX 4090	45-65%	US/EU

Supported Models

Freemium Tier

• LLaMA: 7B, 13B, 70B variants
• GPT-2: Small, Medium, Large, XL
• BERT: Base, Large variants
• Stable Diffusion: v1.5, v2.1
• Mistral: 7B, 7B-Instruct

Paid Tier

• All Freemium models plus:
• Custom model imports
• Fine-tuned variants
• Proprietary models
• Custom architectures

Supported Datasets

Hugging Face (Freemium)

• Computer Vision: COCO, ImageNet-1K, CIFAR-10/100
• NLP: SQuAD, GLUE, WikiText, IMDB
• Multimodal: LAION, Conceptual Captions

Custom Sources (Paid)

• AWS S3 buckets
• Google Cloud Storage
• Azure Blob Storage
• Private data lakes
• Custom APIs and databases

Cost Savings Examples

GPU Type	On-Demand	Spot (AWS)	Spot (RunPod)	Best Price	Savings
A100	$4.06/hr	$1.22/hr	$0.89/hr	$0.84/hr	79%
H100	$7.20/hr	$2.16/hr	$1.59/hr	$1.50/hr	79%
A10G	$1.21/hr	$0.36/hr	$0.27/hr	$0.25/hr	79%

Prices are illustrative and updated in real-time

Requirements

• Terraform >= 1.0
• Cloud provider CLI tools configured
• Python 3.8+ (for query interface)
• Docker (optional, for containerized workloads)

License

MIT License - see LICENSE file for details