Unsloth AI
FreemiumOpen-source no-code web UI for training, running, and exporting open models in a unified local interface.
工具信息
什么是 Unsloth AI？
Unsloth AI is an open-source, no-code local interface that lets you train, run, and export open models completely offline. It provides optimized kernels for faster fine-tuning, multimodal support for text, vision, audio and embeddings, tool calling, automatic dataset creation from documents, and flexible model export—all in one unified web UI that runs on your own hardware.
核心功能
Run models locally
Runs 100% offline on Mac and Windows. Supports GGUF and Safetensors models with tool-calling, web search, and an OpenAI-compatible API. Enables side-by-side model comparison and uploads of images, documents, audio, code files, and more.
No-code training
Auto-creates datasets from PDF, CSV, and JSON documents and starts training with real-time observability. Custom kernels support optimized training for LoRA, FP8, FFT, PT, and 500+ models including text, vision, audio, and embeddings.
Unlimited tool calling and web search
Lets LLMs run unlimited web search and execute Bash and Python. Sandboxes programs so models can test code, generate files, and verify answers with real computation.
Data Recipes
Transforms documents into usable datasets via a graph-node workflow. Uploads unstructured or structured files such as PDFs, CSV, and JSON and automatically converts them into desired formats.
Export models
Exports any model, including fine-tuned models, to safetensors or GGUF for use with llama.cpp, vLLM, Ollama, and more.
Optimized training speed and memory
Provides significantly faster training and lower memory usage than baselines such as FA2, with free open-source fine-tuning claimed at 2x faster on a single NVIDIA GPU and higher tiers claiming up to 30x faster training, 90% less memory, and multi-GPU or multi-node support.
使用场景
- #1
Local offline model inference
Run open models completely offline on personal Mac or Windows devices with multimodal inputs, tool calling, and OpenAI-compatible API access.
- #2
No-code custom model fine-tuning
Train custom models in hours rather than days using auto-generated datasets, optimized kernels, and free Colab or Kaggle notebooks.
- #3
Document-to-dataset conversion
Turn PDFs, CSVs, JSONs, and other files into training-ready datasets via Data Recipes workflows.
- #4
Model export for deployment
Export fine-tuned or base models to formats usable with popular runtimes such as llama.cpp, vLLM, and Ollama.
- #5
Agentic tool-using LLM workflows
Enable models to perform web search, code execution, and computation inside sandboxes to generate files and verify answers.
定价方案
Free open-source standard version (freeware) supports Mistral, Gemma, Llama 1/2/3, 4-bit and 16-bit LoRA; MultiGPU coming soon. Pro provides 2.5x faster training, 20% less VRAM, enhanced MultiGPU up to 8 GPUs (contact for pricing). Enterprise unlocks up to 30x faster training, multi-node support, up to +30% accuracy, 5x faster inference, full training, all Pro features, and customer support (contact for pricing).
替代工具
常见问题
What is Unsloth AI?
Does Unsloth AI work completely offline?
Is there a free version of Unsloth?
Can I create training datasets from my own documents?
Which formats can I export models to?
What platforms does Unsloth support?
Unsloth AI 联系方式与社交媒体
- 公司Unsloth
- 支持
- 网站：
- Twitter
- LinkedIn
- GitHub
- Discord