Unsloth Studio: No-Code Web UI for Training and Running Open Models Locally

Unsloth, the team known for making LLM fine-tuning faster and more efficient, has launched Unsloth Studio (Beta) — an open-source, no-code web UI for training, running, and exporting open models in one unified local interface.

Key Capabilities

• Run GGUF and safetensor models locally on Mac, Windows, and Linux
• Train 500+ models 2x faster with 70% less VRAM (no accuracy loss)
• Support for text, vision, TTS audio, and embedding models
• No dataset needed — auto-create datasets from PDF, CSV, JSON, DOCX, TXT files
• Self-healing tool calling / web search + code execution
• Auto inference parameter tuning and editable chat templates
• Export to GGUF, 16-bit safetensor formats
• Multi-GPU inference and training support
• Model arena for side-by-side comparison

Data Recipes

Powered by NVIDIA DataDesigner, Studio's Data Recipes feature transforms unstructured documents into training-ready datasets through a graph-node workflow. Upload PDFs, CSVs, or JSON files and the system auto-generates synthetic datasets in your desired format.

Models Supported

Fine-tune the latest open models including Qwen3.5 and NVIDIA Nemotron 3, along with hundreds of other text, vision, audio, and embedding models. The system leverages Unsloth's optimized kernels for LoRA, FP8, FFT, and PT training.

Who It's For

Unsloth Studio targets developers and researchers who want the power of model training and inference without the complexity of command-line tools or Python scripts. By keeping everything local, it also addresses privacy concerns for sensitive data.

Source: Unsloth AI | GitHub