Welcome to EzEpoch BETA
Choose how you'd like to start your AI training project
Name Your Project
Give your training setup a descriptive name
🎯 Project Name
Choose a name that describes your training configuration
💡 Suggestions:
Data Processing - New Project
Upload, clean, and prepare your training data
📤 Upload Training Data
Select your training data files (.json, .jsonl, .txt, .csv)
Drag & drop files here
or
Supported formats: .json, .jsonl, .txt, .csv, .tsv ❓
Skip Data Upload
Have a large dataset or prefer to add data later?
Create a package without uploading data now - you'll add your data files to the package after download.
How does this work?
1. Skip to Training Setup →
2. Configure your model and settings →
3. Download package with empty data/ folder →
4. Add your training files to the data/ folder before training
✅ Perfect for large datasets (>1GB)
✅ Saves upload time
✅ Full instructions included in package
📁 Previously Uploaded Data
Select from your previously uploaded files or upload new ones above
Smart Data Optimization
After upload, use Data Analysis & Cleaning to optimize your files for training performance.
- Detects and fixes content issues
- Removes short, unhelpful entries
- Splits overly long content into manageable chunks
- Preserves question-answer structure for training
📤 Uploading to Cloud Storage
Uploading your files securely to cloud storage
🧹 Data Cleaning
Automatic data cleaning and quality assessment
❓Upload data to see cleaning results
🖼️ Link Images (Optional)
For multimodal training, link images to your text data
❓✅ Data Review
Review your processed data before training setup
Process data to see summary
Training Setup - New Project
Configure your AI model and training parameters
All parameters are automatically optimized for your selected model and GPU. Simply choose your model and GPU - the system handles batch size, memory optimization, and compatibility automatically!
🎯 Project Name
Give your training project a descriptive name
🤖 Model Selection
Choose your AI model and training type
🖥️ GPU Configuration
Configure GPU settings and memory optimization
⚙️ Training Parameters
Configure core training settings
- Sorts data by difficulty (easy → hard)
- Model learns basics first, then complex patterns
- 20-40% better convergence & accuracy
- Prevents early overfitting
Requirements - New Project
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!
🔗 PyTorch + Transformers Compatibility
Select the best combination for your setup
🔧 Individual Version Selection
Select specific versions or enable Auto to let the system choose compatible versions
📦 Dependencies & Libraries
Required software packages and versions
Click "Generate Requirements" to create your dependency list.
Advanced Options - New Project
Fine-tune your training with advanced parameters
All settings are pre-optimized for your chosen AI model and GPU. You can proceed through the entire process with these defaults for excellent results, or customize any parameter below for specific needs.
🚀 AI Advanced Setup
Let AI optimize your training parameters automatically
Model Access - New Project
Add your HuggingFace & GitHub tokens for model access
Add your HuggingFace token to access gated models. Tokens are encrypted and stored securely. Click the 👁️ button to show/hide your token safely.
🤗 HuggingFace Token
Your HuggingFace token is used to download and upload models to your account.
Get your token →🐙 GitHub Token (Optional)
GitHub token enables discovery of models from research repositories and increases API rate limits.
Create token →Multi-Repository Support
EzEpoch searches multiple AI model repositories to give you access to the latest models:
- 🤗 HuggingFace Hub - 500K+ models (requires token for gated models)
- 🐙 GitHub Repositories - Research models and implementations (optional token)
- 📄 Papers with Code - State-of-the-art research models (no token needed)
- 🔥 PyTorch Hub - Native PyTorch models (no token needed)
GPT Auto-Analysis: All discovered models get optimized settings generated automatically!
🤖 AI Model Library
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!
🔐 Repository Access Status
Connect to repositories to discover and access their models
HuggingFace Hub
500K+ models, gated access available
GitHub Repositories
Research models and implementations
Papers with Code
State-of-the-art research models
PyTorch Hub
Native PyTorch models
📚 Browse Models
Select a repository and browse available models
Select a repository to view available models...
👤 User Profile
Manage your account information and preferences
Update your profile information, manage your subscription, and configure notification preferences. All changes are saved automatically.
📝 Personal Information
📞 Contact Information
🤖 AI Monitoring Preferences
🔐 API Hash & Sessions
No active sessions
📦 Build Package - New Project
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!
🖥️ Platform Support
Universal package for RunPod, Vast.ai, Lambda Labs, AWS, GCP
📁 Package Contents
| 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 |
🚀 Package Creation
☁️ Cloud GPU Deployment - New Project
Deploy your training job 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.
🔑 Cloud Provider API Keys
📦 Recent Packages (Last 10)
💡 Your last 10 packages will appear here
Build packages in the "Build Package" tab, then click Refresh to see them here
🖥️ Available GPUs
📋 Your Requirements (from Setup tab)
💰 Sorted by lowest price — Best deals first! (Uncheck "Match Requirements" to see all options)
Connect your API keys above to browse available GPUs
📊 Training Dashboard
Monitor your active training sessions in real-time
📁 Your Training Sessions
💳 Subscription & Billing
Manage your plan, sessions, and billing
Pay only when training actually starts, not during package creation. Automatic refunds for failures under 15 minutes. Sessions rollover at 50% for paid plans.
📊 Current Plan
📈 Recent Training Activity
Your recent training sessions and resource usage
🧮 How Sessions Work - Model Size Matters
Larger models require more GPT monitoring calls, so they use more sessions
0-20B Models
Mistral 7B, Llama 3.2 8B, Qwen 14B, etc.
21-70B Models
Llama 3.1 70B, Falcon 40B, Mixtral 8x7B, etc.
71-200B Models
Llama 3.2 90B, BLOOM 176B, Falcon 180B, etc.
200B+ Models
Llama 3.1 405B, GPT-3 scale models, etc.
💡 Why Session Multipliers?
Larger models generate more training logs and require more frequent GPT analysis (every 100 steps). A 7B model might train for 4 hours, while a 200B model can run for 40+ hours. More monitoring = more GPT API costs = more sessions used. This keeps pricing fair and aligned with actual resource usage.
📊 Example Calculations:
- Starter Plan (5 sessions): Train 5x 7B models OR 2x 70B models OR 1x 200B model
- Developer Plan (20 sessions): Train 20x 7B models OR 10x 70B models OR 5x 200B models
- Professional Plan (50 sessions): Train 50x 7B models OR 25x 70B models OR 12x 200B models
🚀 Available Plans
Choose the plan that fits your training needs
🛒 Add-On Products & One-Time Purchases
Enhance your workflow with standalone tools or buy sessions as needed
🛡️ AI Training Insurance - Zero-Loss Guarantee
We turn 40% industry failure rates into 99% success rates. Guaranteed.
✅ Our 99% Success Guarantee
Use EzEpoch's recommended settings and let our AI Guardian monitor your training. If it fails, we refund you. Period.
✅ Covered - Full Refund
AI Guardian Failed
Used recommended settings but training crashed and our AI didn't recover it? Full refund.
Setup Configuration Errors
Package won't create, requirements wrong, or settings miscalculated? Full refund.
Platform Errors
Bootstrap auth fails, dashboard doesn't connect, or monitoring stops working? Full refund.
First Training Extended Protection
Your first training gets extra protection (30 minutes) while you learn the platform.
❌ Not Covered - No Automatic Refund
Custom Settings Override
Changed batch size, learning rate, or other settings manually? You declined our AI protection.
Data Quality Issues
Training failed due to corrupted data, wrong format, or insufficient data? That's your data, not our platform.
GPU Provider Issues
RunPod crashed? Vast.ai down? GPU out of memory with your custom data? That's between you and your GPU provider.
User Stopped Training
You clicked stop/pause and decided not to continue? That's not a platform failure.
💡 We Want You To Succeed!
🔄 Crash Recovery: Our AI Guardian automatically analyzes crashes and restarts training with corrected settings. Resumes from last checkpoint - no progress lost!
🔁 Unlimited Restarts: Each session includes unlimited restart attempts. Training keeps going until YOU succeed.
🎯 Expert Support: Stuck? Our team will help debug your setup, review your settings, and guide you to success.
📊 Dashboard Control: Monitor live metrics, adjust settings on-the-fly, and let GPT optimize every 100 steps automatically.
🦶 So Easy, Bigfoot Can Do It!
Industry Standard: 40% of AI training jobs fail due to configuration errors, memory issues, and instability.
With EzEpoch: 99% success rate because our AI calculates perfect settings, monitors every 100 steps, and auto-recovers from crashes.
Think of it like insurance: We insure your SETUP and MONITORING. You provide the GPU and data. Follow our recommended settings → We guarantee success. Override our settings → You're on your own (but we'll still help!).
💰 Request Refund
Training failed using our recommended settings? Request your refund below.
ℹ️ Refund Requirements
- Must have used EzEpoch's recommended settings (no custom overrides)
- Platform error must be verified (setup, monitoring, or dashboard failure)
- GPU provider issues are not covered (contact your GPU provider)
- Data quality issues are not covered (we'll help you fix your data)
- First training gets extended grace period automatically
📚 Complete EzEpoch Guide
Everything you need to know about AI training with smart recommendations
🚀 Quick Start - 5 Minutes to Training
Follow these steps to create your first training package
1️⃣ Choose Your Model
Select from 100+ pre-configured AI models. The system automatically detects model size and requirements.
2️⃣ Select GPU
Look for ⭐ stars - these GPUs can fully train your model!
3️⃣ Configure Training
The Advanced tab automatically selects optimal settings. Change precision, optimizer, or training method if needed.
4️⃣ Generate Package
Click "Create Package" - system validates everything, resolves conflicts, and creates a ready-to-run training package!
⭐ GPU Star System
Instantly see which GPUs can handle your model
How It Works
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!
✅ With Star (⭐)
This GPU has enough memory to fully train your model with any optimizer and settings. Full control!
ℹ️ Without Star
GPU needs LoRA or QLoRA for this model. You'll see a message with recommendations and alternative GPUs.
Example: Mistral 7B
⭐ H100 (80GB) ← Can fully train!
⭐ A100 (80GB) ← Can fully train!
L40S (48GB) ← LoRA/QLoRA recommended
RTX 4090 (24GB) ← LoRA/QLoRA recommended
L40S (48GB) requires LoRA or QLoRA for Mistral-7B
Full training needs 86GB
Recommended: A100 (80GB), H100 (80GB), H200 (141GB)
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.
🧠 Smart Optimizer System
Automatic compatibility checking and recommendations
What It Does
The Smart Optimizer analyzes your model + GPU combination and:
- ✅ Enables optimizers that will work
- ❌ Greys out optimizers that won't fit in memory
- ⭐ Recommends the best optimizer for your setup
- 📊 Shows memory usage for each option
Optimizer Comparison
| 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!
📋 Requirements
Generate dependencies and check for conflicts
Key Features:
- ✓ PyTorch + Transformers + CUDA compatibility presets
- ✓ Auto-Select preset chooses optimal versions for your GPU
- ✓ Automatic dependency generation (66+ optimized packages)
- ✓ Conflict-free installation with proper ordering
- ✓ Latest, Stable, Older, Legacy, and CPU-only presets available
💡 Tip: Use "Auto-Select (Recommended)" preset for best compatibility. It automatically picks the right PyTorch, Transformers, and CUDA versions for your selected GPU.
🔑 API Keys
Set up authentication for external services
Key Features:
- ✓ HuggingFace token for model access (required for gated models)
- ✓ GitHub token for research repositories (optional, increases rate limits)
- ✓ Eyeball button (👁️) to show/hide tokens securely
- ✓ Test & Save validates tokens before storing
- ✓ Multi-repository support: HuggingFace, GitHub, Papers with Code, PyTorch Hub
💡 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.
📦 Build Package
Create complete training packages for deployment
Key Features:
- ✓ Universal platform support (RunPod, Vast.ai, Lambda Labs, AWS, GCP)
- ✓ Package contents preview (requirements.txt, main.py, all.env, setup.sh, README.md)
- ✓ Custom package naming (defaults to project name)
- ✓ One-click package creation
- ✓ Core files streamed from server for updates and flexibility
💡 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.
🤗 AI Models
Browse and select from thousands of AI models
Key Features:
- ✓ Browse 100,000+ models from HuggingFace Hub
- ✓ Text, Vision, and Audio models organized by category
- ✓ Repository access status (HuggingFace, GitHub, Papers with Code, PyTorch Hub)
- ✓ Model details: size, type, description
- ✓ "Add to Setup" button instantly configures the selected model
💡 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.
🤖 AI-Powered Dashboard
Revolutionary AI monitoring that sets EzEpoch apart
GPT monitors your training and provides real-time insights
🚀 Revolutionary AI Features:
- 🤖 GPT Training Analysis: AI continuously monitors training progress and identifies issues
- 💡 Smart Recommendations: Real-time suggestions for parameter adjustments
- ⚠️ Predictive Alerts: AI predicts and prevents training failures before they happen
- 📈 Intelligent Optimization: Automatic parameter tuning based on training patterns
- 🎯 Performance Insights: AI explains why certain settings work better
- 🔄 Adaptive Learning: System learns from your training patterns to improve future sessions
🌟 EzEpoch's Competitive Edge: We're the ONLY platform that uses GPT 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!
👤 Profile
Manage your account and preferences
Key Features:
- ✓ Personal information and contact details management
- ✓ AI monitoring preferences (email notifications, frequency)
- ✓ Your unique API hash for authentication
- ✓ Active training sessions display
- ✓ Timezone configuration for accurate scheduling
💡 Tip: Your API hash is used for secure authentication across all EzEpoch services. Copy it for use in your training packages or dashboard access.
📖 Complete Settings Dictionary
Comprehensive guide to every parameter and setting
🎯 Training Parameters
⚡ Precision & Optimization
🧠 Optimizers
🎛️ Training Methods
🖥️ GPU & Memory
🤖 AI Monitoring (EzEpoch Exclusive)
🔧 MyEzSetup - Dependency Checker
Intelligent dependency management and conflict resolution
🎯 Key Features:
- ✓ Dependency Compatibility Checker: Paste your requirements.txt and get instant compatibility analysis
- ✓ Auto-Generated Pinned Files: Creates conflict-free requirements.txt with proper version pinning
- ✓ Installation Scripts: Generates platform-specific setup scripts (install.sh / install.bat)
- ✓ CUDA Support: CPU-only, CUDA 11.8, and CUDA 12.1 configurations
- ✓ Platform Detection: Auto-detects Windows, Linux, macOS requirements
- ✓ API Access: RESTful API for integration with training pipelines
🌐 Access MyEzSetup: Visit www.myezsetup.com or ezsetup.ezepoch.com
💡 How It Works:
- Paste your requirements.txt file
- Run compatibility check to identify conflicts
- Select target platform and CUDA version
- Generate pinned requirements.txt and installation script
- Copy the script and run it on your system
🎉 Free Tier: Sign up for free and get 5 dependency checks per month. Upgrade to MyEzSetup ($4.99/month) or MyEzSetup API ($9.99/month) for unlimited access.