H2O LLM Studio
An open-source LLM fine-tuning tool with a browser interface. Runs on Ubuntu with NVIDIA GPUs or in Docker, under the Apache 2.0 license.
H2O LLM Studio is a self-hosted tool for teams that want to adapt language models to their own datasets without writing training code. Its browser interface brings training experiments, evaluation, and model testing into one place. The project is open source under Apache 2.0.
It runs on Ubuntu with a recent NVIDIA GPU and also supports Docker. Larger models may need substantial GPU memory; the project recommends at least 24 GB for them. You can run training on your own hardware or on a cloud GPU machine such as RunPod. Multiple GPUs are supported.
LoRA and 8-bit training reduce the memory needed to fine-tune models. The interface lets you adjust training settings, monitor experiments, and compare model performance visually. Evaluation metrics help assess generated answers, while built-in chat lets you test a trained model directly. Weights & Biases integration provides another way to track experiments.
The tool covers instruction and chat fine-tuning, plus classification and regression tasks with small language models. It also supports DPO, IPO, and KTO methods for aligning model responses with preferences. These capabilities suit teams building models for a specific business task, including smaller models intended for mobile or offline applications.
A command-line interface supports work outside the GUI. You can export trained models to Hugging Face Hub for sharing.
