EzEpoch is a fully automated AI training engine that deploys training in minutes,
auto-configures every model family, generates all required packages, detects true MSL,
structures data, deploys to GPU clouds, monitors training with AI Guardian,
recovers from crashes, preserves checkpoints, and exports full or quantized models.
Choose Any Model — Or Bring Your Own
Included text, vision, and audio models
Create bimodal, trimodal, or all-in-one multimodal models
Import from Hugging Face, GitHub, Papers With Code, or Transformers repos
Custom models auto-configure and stay saved with a ⭐
Automatic Configuration
EzEpoch fills in architecture defaults, tokenizer rules, optimizer/scheduler settings,
LoRA/QLoRA/full-finetune options, multimodal routing, preprocessing, and expert-level parameters.
Automatic Package Generation
EzEpoch generates all required packages, installs them in the correct order, avoids conflicts,
and builds the full training environment.
One-Click GPU Deployment
With Vast.ai or RunPod API keys, EzEpoch deploys the GPU instance, pushes your build package,
sends your dataset directly from your computer to the GPU, and starts the full workflow automatically.
Pre-Training Intelligence
True MSL Detection — finds the real max sequence length
Optional Data Cleaning — removes broken samples and normalizes formatting
Curriculum + Batch Building — structures data based on your config
Training With AI Guardian
AI Guardian monitors training, adjusts parameters, prevents instability, detects anomalies,
recovers from crashes, analyzes logs, fixes issues, restarts training, and preserves checkpoints.
Export Your Model
Download full model
Quantize + download
Push to Hugging Face
Generate a training report
Setup Requirements (Only 3 Things)
Hugging Face Token — fine-grained token with 4 required permissions
Vast.ai or RunPod API Key — for GPU deployment
GitHub Token (Optional) — for private repos or custom model code
