Fine-tuning lab on OpenGPU Network

Train open models on a living reef of GPUs

Upload a dataset, pick a base model, launch LoRA fine-tuning. No cluster setup. No YAML. Download your adapter when the job completes.

openreef.network · pay per job · CUDA & ROCm aware

Four steps. Then an adapter.

The dashboard is the lab bench. The reef is the compute underneath.

JSONL, CSV or plain text. We normalize and validate before you spend a credit.

1.5B–70B open models. Fast, balanced or quality. LoRA with hardware-aware defaults.

SmartRoute finds capacity on the OpenGPU network — NVIDIA or AMD, no cloud account.

Training finishes, you get a secure download. Pay only what the job actually used.

A four-step lab flow instead of Docker, Axolotl configs and cluster scripts.

Jobs run on OpenGPU providers. Local mode for your own card when you want it.

See price before confirm. Partial refunds if a job fails. No monthly GPU tax.

Create an account, top up credits, and launch your first job in minutes.