H2O LLM Studio
A framework and no-code GUI for fine-tuning large language models, with a command-line interface for the same workflows.
H2O LLM Studio lets a user fine-tune an LLM with a wide range of hyperparameters through a graphical interface built specifically for LLMs, without writing code.
It supports LoRA and 8-bit training for a reduced memory footprint, an experimental reinforcement-learning fine-tuning mode, advanced evaluation metrics for judging generated answers, visual tracking and comparison of model performance with optional Weights & Biases integration, an in-app chat for instant feedback, and export to the Hugging Face Hub. Recent changes add a causal-regression-modeling problem type, move from RLHF toward DPO, IPO and KTO optimization, and add a paired-preference loss for training on manually matched positive and negative examples.
It runs through its GUI, Docker, or a CLI; a cloud quickstart uses a runpod.io instance. It is written in Python and licensed under Apache-2.0.