← Back to work
Forge — Model Trainer
Product

Forge — Model Trainer

A desktop app for fine-tuning custom models on rented GPUs — launch a training run without touching a terminal.

Overview

Forge is the Strata Model Trainer: an Electron + Next.js desktop application that turns fine-tuning a custom LLM into a guided workflow. It provisions GPUs on vast.ai, manages datasets and hyperparameters, launches and monitors training runs, and pulls the finished model back — the toolchain behind ConstructionAI, packaged so it can be repeated for any model.

The Challenge

Fine-tuning a model usually means a tangle of SSH sessions, rented GPUs, and shell scripts that only the person who wrote them can run.

The Solution

Wrap the whole loop — GPU provisioning, dataset prep, hyperparameters, launch, and monitoring — in a desktop GUI on top of vast.ai, so a training run is a few clicks instead of a terminal marathon.

Key features

One-click GPU provisioning on vast.ai
Dataset and hyperparameter management
Launch and monitor training runs from a GUI
Electron + Next.js desktop app
The pipeline behind ConstructionAI, productized