Developer builds fine-tuning studio that runs a 1.7B model on just 3.2 GB VRAM

A GenAI engineer named Pranjul Rathour has built FineTune Studio, an open-source tool that lets users fine-tune the Qwen3-1.7B language model using as little as 3.2 GB of VRAM via QLoRA. The project was created to address the gap in fine-tuning tutorials, which typically assume access to expensive GPU hardware like an A100. The platform features dataset validation on upload, live streaming of training metrics, and a side-by-side evaluation page that compares the base and fine-tuned model on identical prompts. It includes 17 backend endpoints, 107 passing tests, and supports three inference paths — local, vLLM, or a Hugging Face Space. The source code and full architecture are publicly available on GitHub.