Train your own models.
No infra headaches.
Easy Training is a new way to specialize models on your own tasks, your own data, and your own quality bar. Upload a dataset, pick a base model, and let Compute handle the rest.
How Easy Training works.
For teams that want full control over model behavior without managing training infrastructure.
01 Upload
Bring your own dataset. JSON, CSV, or conversational logs.
02 Select
Choose a base model from Hugging Face or your own fine-tuned checkpoint.
03 Train
Our AI agent helps you find the perfect dataset and train a specialized model through knowledge distillation.
04 Deploy
Trained models become available behind your own OpenAI-compatible endpoint in Compute.
An AI agent specialized in training.
Easy Training includes a dedicated AI agent that understands your task and guides the entire training process—finding the right data, choosing the right approach, and ensuring the resulting model does exactly what you need.
Dataset discovery
Not sure what data to train on? Describe your task and the agent searches for relevant datasets—public, curated, or synthetic—that will make your model specialize in exactly what you need.
Knowledge distillation
Want a small, fast model for a specific task? The agent can guide you through training a specialized student model by distilling knowledge from a larger teacher model—keeping costs low while preserving the capability you need.
Example workflow
Describe your task: "I need a model that classifies support tickets into billing, technical, and general categories."
Agent suggests a dataset: Curated public dataset of support conversations, plus any existing logs you can upload.
Agent proposes distillation: Use a larger teacher model to train a smaller, specialized student that handles your task at a fraction of the inference cost.
Why Easy Training?
Most teams train models because the base model does not understand their domain well enough. The prompt engineering ceiling is real—fine-tuning on your own data closes that gap.
Easy Training removes the infrastructure burden: no GPU clusters to manage, no training scripts to debug, no checkpoint management to figure out. The trained model lands directly behind your OpenAI-compatible endpoint in Compute, ready to serve production traffic.
Domain specialization
Train on support tickets, legal documents, code repositories, or customer conversations. The model learns your language, not generic web text.
Consistent behavior
Prompt engineering varies. A fine-tuned model trained on your data produces consistent outputs regardless of how you frame the prompt.
Lower inference cost
A smaller specialized model often outperforms a large general-purpose model on narrow tasks. Easy Training lets you right-size the model to your task.
Full ownership
Trained models are yours. Deploy behind your balancer, route traffic through Smart Balancers, apply firewall policies—just like any other model in Compute.
Be the first to know when Easy Training launches.
We are building fast. If you have a specific training need or want early access, let us know.