Getting Started with Ertas: Fine-Tune and Deploy Custom AI Models

Most AI platforms ask you to pay per token and hope for the best with generic models. Ertas works differently. You fine-tune models through a visual interface using cloud GPUs, then download and run them on your own hardware — or deploy via Ertas Cloud when it launches.

This guide walks you through the three core steps: uploading a dataset, fine-tuning in Studio, and downloading your finished model.

Step 1: Upload Your Dataset

Ertas accepts training data in JSONL format — one JSON object per line, which is the standard for fine-tuning large language models. If you don't have a dataset ready, you can import one directly from Hugging Face using a URL.

Preparing Your Data

A typical JSONL training file looks like this:

{"prompt": "Summarize this support ticket", "completion": "Customer reported billing issue..."}
{"prompt": "Classify this email", "completion": "Category: Feature Request"}

Each line is a self-contained training example. Ertas validates your file on upload and flags formatting issues before you start a fine-tuning run, so you catch problems early.

Don't have training data yet? Browse Hugging Face's open datasets and paste the URL directly into the Ertas upload interface. It handles the download and conversion for you.

Step 2: Fine-Tune in Studio

Once your dataset is uploaded, you move into Ertas Studio — a canvas-driven visual environment where you configure and launch fine-tuning jobs on cloud GPUs.

What Makes Studio Different

• Multiple models at once — Fine-tune several base models simultaneously on the same dataset, then compare results side by side
• Saved knowledge — Ertas preserves the knowledge from each fine-tuning run, so you can iterate and test across different use cases without starting from scratch
• Visual workflow — No command-line scripts or YAML configs. Configure hyperparameters, select base models, and monitor training progress through the web interface
• Cloud GPUs — Fine-tuning runs on fast cloud hardware, so you don't need expensive local GPUs

Studio is designed so that you spend time evaluating results, not debugging training pipelines.

Step 3: Download and Deploy

When your model is ready, download it as a GGUF file — the widely-supported format for running large language models on consumer hardware. GGUF models work with tools like llama.cpp, Ollama, LM Studio, and many others. Cloud deployment via Ertas Cloud is coming soon.

Why Custom Models Matter

• No recurring API costs — Run inference as many times as you want on your own hardware
• Full data privacy at inference — Queries and responses never touch an external server when running locally
• No vendor lock-in — GGUF is an open format. Your model works with any compatible runtime
• Domain expertise baked in — A fine-tuned model understands your product better than any amount of prompt engineering

What's on the Roadmap

Ertas is building toward a full ecosystem for custom AI:

1. Studio (In Development) — The cloud fine-tuning interface described above, with smart data synthesis suggestions coming soon
2. Hub (Coming Next) — Discover and share fine-tuned models with the community
3. Cloud (On the Horizon) — Deploy fine-tuned models as API endpoints with workflow automation integration
4. Vault (Enterprise) — Enterprise-grade encrypted storage for datasets and secrets

Join the Waitlist

Ertas is currently in development. Sign up for the waitlist on our homepage to get early access and help shape the platform.

---

Build models you trust. Train them on your data. Deploy them on your terms.