#2929th of 60 in Training and fine-tuning
h2o-llmstudio
No-code GUI and CLI for fine-tuning LLMs with LoRA and DPO
- Stars
- 5.2k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Last release
- Nov 2026
- Language
- Python
4,869 on 21 Apr 20265,195 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
H2O LLM Studio is a Python framework with an H2O Wave web GUI and a CLI for fine-tuning LLMs on your own datasets. It supports LoRA, 8-bit training, DPO/IPO/KTO optimization, and classification and regression problem types, with DeepSpeed or DDP for multi-GPU runs. Experiments can be tracked and compared in the UI, chatted with, and exported to the Hugging Face Hub.
Who it is for: Teams fine-tuning open LLMs on their own NVIDIA GPUs, with or without code
Strengths
- GUI covers dataset import, experiment tracking, comparison, chat and Hugging Face export
- Supports LoRA, 8-bit training, DPO/IPO/KTO, plus causal classification and regression
- Multi-GPU via DDP or DeepSpeed; YAML-configured CLI for scripted runs
- Official Docker image on Docker Hub and a ready RunPod template
Weaknesses
- Requires an NVIDIA GPU with drivers >= 580; no CPU or non-NVIDIA path documented
- Backwards compatibility not guaranteed; README advises pinning versions
- DeepSpeed needs NVLink and a system CUDA Toolkit install
- RLHF was removed; reinforcement learning is limited to DPO/IPO/KTO
What it needs
- GPU required
- Docker
- Needs NVIDIA drivers >= 580, CUDA 13 (PyTorch), H2O Wave, uv, Python 3.12, nvidia-container-toolkit (Docker)
- Models: Hugging Face LLMs, LoRA, 8-bit training, DPO/IPO/KTO
- port 10101
What changed
- v1.15.4 ↗ (opens in a new tab)Security fix
CVE-2026-111497. “Move to NumPy 2 and bump pyarrow to 26.0.1 to fix CVE-2026-111497 by @pascal-pfeiffer in https://github.com/h2oai/h2o-llmstudio/pull/1135”
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Training and fine-tuning
See all 60| Rank | Project | Score |
|---|---|---|
| 1 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 89 out of 100 |
| 2 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 3 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 87 out of 100 |
| #4 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 86 out of 100 |
| #5 | megatron-lmDistributed training library for large transformer models on NVIDIA GPUs | 85 out of 100 |