Configure your AI model training in one place
Make your core choices - we'll handle the optimization
PyTorch + Transformers Compatibility & Package Management
Just click "Generate Requirements" and we'll create 66+ optimized packages with perfect compatibility. All conflicts are automatically resolved and installation order is optimized!
Select the best combination for your setup
Required software packages and versions
Click "Generate Requirements" to create your dependency list.
Manage all your API keys, tokens, and SSH access in one place
All API keys are encrypted and stored securely. Configure them once here, and they'll sync automatically to the Wizard and Cloud Deploy sections.
Your HuggingFace token is used to download and upload models to your account.
Get your token →GitHub token enables discovery of models from research repositories and increases API rate limits.
Create token →Your Vast.ai API key is used to browse and rent GPUs from their marketplace.
Get your API key →Your RunPod API key is used to create and manage GPU pods.
Get your API key →🔑 Need to generate an SSH key?
Your SSH public key enables secure, passwordless access to your cloud GPU instances.
⚠️ IMPORTANT: Copy the PUBLIC key (.pub file), NOT the private key!
1️⃣ Generate a new SSH key (if you don't have one):
ssh-keygen -t rsa -b 4096
Press Enter 3 times (accept defaults, no passphrase needed for cloud GPUs)
2️⃣ Display your PUBLIC key:
Windows (PowerShell/CMD):
type %USERPROFILE%\.ssh\id_rsa.pub
Mac/Linux:
cat ~/.ssh/id_rsa.pub
3️⃣ Your key should look like this:
ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAACAQ... jason@jasons-laptop
✅ Copy the ENTIRE line (including the comment at the end)
4️⃣ Add your SSH key to cloud providers:
⚠️ You must add your SSH key to both providers before deploying!
Add your own GPU servers (local machines, AWS, GCP, etc.) for training. These will appear in the GPU list under "My Custom SSH".
Configure your API keys once here, and they'll automatically populate in:
Security: All keys are encrypted at rest and in transit. We never log or expose your API keys.
EzEpoch searches multiple AI model repositories to give you access to the latest models:
AI Auto-Analysis: All discovered models get optimized settings generated automatically!
Browse, search, and manage AI models from HuggingFace
Browse and search from the world's largest AI model library. Use filters to find models compatible with your GPU and training goals. Your favorites and recent models are automatically saved!
Connect to repositories to discover and access their models
500K+ models, gated access available
Research models and implementations
State-of-the-art research models
Native PyTorch models
Select a repository and browse available models
Select a repository to view available models...
Manage your account information and preferences
Update your profile information, manage your subscription, and configure notification preferences. All changes are saved automatically.
Create a complete training package for cloud deployment
Everything is ready! Just click "Create Package" and we'll generate a complete training package with all dependencies, scripts, monitoring tools, and your data folders. Ready for any cloud GPU platform!
| File | Status | Size | Description |
|---|---|---|---|
| requirements.txt | ⏳ Pending | ~20KB | Python dependencies (~22 selective packages, EzSetup validated) |
| main.py | ✅ Complete | ~5KB | Primary training script with repo-specific model loading |
| all.env | ✅ Complete | ~600B | API keys, model repos, and training configuration |
| setup.sh | ✅ Complete | ~750B | Environment setup script with EzSetup ordering + venv creation |
| README.md | ✅ Complete | ~1KB | Setup instructions and quick-start guide for RunPod/Vast.ai |
Select a package and deploy to Vast.ai or RunPod with one click
Connect your Vast.ai or RunPod account, browse available GPUs, and deploy your training package directly to the cloud.
Measure real memory savings from GCP, Flash Attention, and optimizers
Searches marketplace for best matching GPU
Creates pod with your exact specifications
Connect your API keys above to browse available GPUs
Monitor your active training sessions in real-time
Manage your plan, sessions, and billing
All paid plans include unlimited training sessions. Train as much as you need - we don't host GPUs, you do. Our job is to make sure your training works the first time.
Train unlimited with confidence - one saved failure pays for months of subscription
Choose the plan that fits your training needs
Enhance your workflow with standalone tools or buy sessions as needed
Unlimited training means unlimited support - we're here to help you every step of the way
AI Guardian automatically analyzes crashes and restarts training with corrected settings. Resumes from last checkpoint - no progress lost!
Training keeps going until YOU succeed. No limits on restart attempts - our goal is your successful model.
Stuck? Our team will help debug your setup, review your settings, and guide you to success.
Monitor live metrics, adjust settings on-the-fly, and let AI Guardian optimize every 100 steps automatically.
Industry Standard: 40% of AI training jobs fail. With EzEpoch: 99% success rate because our AI calculates perfect settings and auto-recovers from crashes.
Everything you need to know about AI training with smart recommendations
Follow these steps to create your first training package
Select from 100+ pre-configured AI models. The system automatically detects model size and requirements.
Look for ⭐ stars - these GPUs can fully train your model!
The Advanced tab automatically selects optimal settings. Change precision, optimizer, or training method if needed.
Click "Create Package" - system validates everything, resolves conflicts, and creates a ready-to-run training package!
Instantly see which GPUs can handle your model
When you select a model, the system calculates memory requirements for full fine-tuning with AdamW optimizer (the industry standard). GPUs with enough memory get a ⭐ star!
This GPU has enough memory to fully train your model with any optimizer and settings. Full control!
GPU needs LoRA or QLoRA for this model. You'll see a message with recommendations and alternative GPUs.
The star system is based on full fine-tuning. LoRA/QLoRA can work on smaller GPUs with 90-95% of full training quality! Don't be afraid to use them.
Automatic compatibility checking and recommendations
The Smart Optimizer analyzes your model + GPU combination and:
| Optimizer | Memory | Quality | Best For |
|---|---|---|---|
| AdamW | Very High | 100% | Industry standard, best quality |
| 8-bit Adam ⭐ | Low | 99% | Recommended! 87% less memory, same quality |
| Lion | Medium | 101% | Often better than AdamW, less memory |
| Adafactor | Very Low | 95% | Large models, memory-constrained |
| SGD | Medium | 85-90% | Simple tasks, experimentation |
System automatically selects 8-bit Adam for most setups - it's the perfect balance of memory efficiency and quality!
Generate dependencies and check for conflicts
💡 Tip: Use "Auto-Select (Recommended)" preset for best compatibility. It automatically picks the right PyTorch, Transformers, and CUDA versions for your selected GPU.
Set up authentication for external services
💡 Tip: Click the 👁️ button to toggle token visibility while entering. All tokens are encrypted and securely stored. Only HuggingFace token is required for most models.
Create complete training packages for deployment
💡 Tip: After creating your package, download the ZIP and upload it to your preferred GPU provider. Most files (like main.py, requirements.txt) are streamed from our server during bootstrap.sh execution, allowing us to provide updates without re-downloading.
Browse and select from thousands of AI models
💡 Tip: Filter models by type (Text, Vision, Audio) and use the search bar to find specific models. All models include AI-optimized settings automatically generated for your GPU.
Revolutionary AI monitoring that sets EzEpoch apart
🌟 EzEpoch's Competitive Edge: We're the ONLY platform that uses custom AI to actively monitor and optimize your training in real-time. While others just show metrics, we provide intelligent insights that save you time, money, and prevent costly failures!
Manage your account and preferences
💡 Tip: Your API hash is used for secure authentication across all EzEpoch services. Copy it for use in your training packages or dashboard access.
Comprehensive guide to every parameter and setting
Prepare your training data locally. No cloud uploads, complete privacy.
💻 Desktop Application: DataLab Pro runs entirely on your computer. Your data never leaves your machine.
📥 Download: Available as a Windows .exe or Python script for Mac/Linux. Included FREE with Developer+ plans.
Stop wasting hours on Python dependency hell. EzSetup fixes it in seconds.
🌐 Access EzSetup: Visit ezsetup-api.herokuapp.com
💡 How It Works:
🎉 Free Tier: 5 dependency checks per month. Upgrade to EzSetup ($4.99/month) or EzSetup API ($9.99/month) for unlimited access.