Use Case

Fine-Tuning

Fine-tuning adapts a pre-trained foundation model to a specific domain, task, or style using a smaller labelled dataset. Parameter-efficient methods such as LoRA and QLoRA dramatically reduce VRAM requirements compared to full fine-tuning, making mid-range GPUs viable for many jobs. Cost-per-run is typically lower than pre-training, so spot or interruptible instances are often a practical choice. This page links the GPUs and providers best suited to fine-tuning workloads across a range of model sizes and budgets.