QTune is a comprehensive web application for fine-tuning language models on consumer GPUs with as little as 8GB of VRAM. Contact developer in Telegram: https://t.me/xkcd0000 Telegram channel: https://t.me/curseknowledge
Hey everyone! I’m Ruslan, the creator of QTune — a web app built with Gradio to simplify the fine-tuning of language models right on your local machine. It supports QLoRA-based workflows on GPUs with as little as 8 GB VRAM, integrates with Hugging Face, OpenRouter (for dataset generation), and even offers model conversion to GGUF format or direct deployment via Ollama.
QTune provides an intuitive, end-to-end pipeline: choose your model, prepare or upload a dataset, configure training parameters, and monitor live training logs—all from your browser. I built this because I wanted something accessible, transparent, and efficient for developers, educators, and hobbyists alike.
I’d love to hear your feedback, ideas, or feature requests. Happy to help if you run into any issues setting it up—just drop a note here!
Thanks for checking out QTune, and I hope it makes LLM fine-tuning more approachable for everyone.
Report
This library seems straightforward at first glance, but it tackles a significant cost/performance challenge. Much respect!
Report
Any chance you’ll support adapters like LoRA fusion right out of the box? That would be amazing.
Report
I’m not usually a fan of Gradio UIs — they can feel clunky after a while. Maybe a React dashboard could be in the works for the future?
Report
Wow, this is just what I needed! I’ve been a bit hesitant to dive into finetuning, but this looks totally manageable! 👏
Report
As someone new to ML, I’ve always felt lost in the command line chaos… having a browser-based UI is such a breath of fresh air.
Report
I’m really impressed with the Hugging Face + OpenRouter integration. Dataset generation is usually the biggest headache.
This library seems straightforward at first glance, but it tackles a significant cost/performance challenge. Much respect!
Any chance you’ll support adapters like LoRA fusion right out of the box? That would be amazing.
I’m not usually a fan of Gradio UIs — they can feel clunky after a while. Maybe a React dashboard could be in the works for the future?
Wow, this is just what I needed! I’ve been a bit hesitant to dive into finetuning, but this looks totally manageable! 👏
As someone new to ML, I’ve always felt lost in the command line chaos… having a browser-based UI is such a breath of fresh air.
I’m really impressed with the Hugging Face + OpenRouter integration. Dataset generation is usually the biggest headache.