Skip to main content
Train is BenchGen’s fine-tuning module: a configuration-driven way to improve a model on your specific task without managing GPU infrastructure yourself. Bring a dataset and a base model, configure a LoRA run, and Train handles the rest, handing back a model ready to merge, run, and evaluate.

Bring your data and model

Add a Dataset

Register a training dataset by importing it from HuggingFace or uploading your own file.

Upload a Model

Bring your own base model into Train: which formats can be fine-tuned and what the archive should contain.

Train and use the result

Fine-tune a Model

Configure and launch a LoRA fine-tuning run, then monitor it to completion.

RLVR Training

Train with GRPO against a verifiable reward, no labeled dataset required.

Merge & Save a Model

Merge a trained LoRA adapter into its base model and push it to BenchGen.

Run Inference

Serve a model you trained as a live, OpenAI-compatible endpoint.

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.
Last modified on September 14, 2026