Train Overview

Train is BenchGen’s fine-tuning module. It gives you a simple, configuration-driven way to improve a model on your specific task — without managing infrastructure.

What Train Does

Upload a dataset, choose a base model, configure a LoRA adapter, and Train handles the rest. When the run completes you download a merged model ready for deployment in Agents or evaluation in Eval. You get:

Managed fine-tuning runs (no GPU provisioning)
LoRA adapter training with configurable hyperparameters
One-click adapter merging
Inference endpoint for immediate testing

When to Use Train

Situation	What to do
Eval shows a cluster of failures	Export those cases and fine-tune on them
You have domain-specific data to improve quality	Upload a custom dataset and run a targeted fine-tune
A merged model is deployed but drifting	Re-run training with updated data
You want to test a smaller, cheaper model	Fine-tune a compact base model on your task

What Train Hands Off

→ Eval: run a benchmark against the fine-tuned model to measure improvement.
→ Agents: connect the merged model as the LLM inside an agent.

Next Steps

Last modified on June 26, 2026