unslothWeb UI for training and running open models like Gemma 4, Qwen3.5, DeepSeek, gpt-oss locally.

unsloth

Unsloth is a web UI and Python framework for training and running open-source language models like Gemma 4, Qwen3.5, DeepSeek, and gpt-oss locally on your machine. It delivers 2x faster training with up to 70% less VRAM while maintaining full model accuracy, making advanced AI capabilities accessible without expensive cloud infrastructure.

Key Features

• Local Model Training: Train 500+ open models up to 2x faster with 70% less VRAM using optimized Triton kernels and mathematical optimizations
• Web-Based Studio UI: Run inference, fine-tune models, manage datasets, and monitor training through an intuitive interface across Windows, Linux, macOS, and WSL
• Multi-Modal Support: Work with text, audio, vision, embedding models, plus tool calling, code execution, and web search capabilities
• Flexible Training Options: Support for full fine-tuning, reinforcement learning, pretraining, 4-bit, 16-bit, and FP8 training with multi-GPU capabilities
• Data Recipe Builder: Automatically create and edit datasets from PDF, CSV, DOCX and other file types using a visual workflow

Use Cases

• Model Fine-tuning: Adapt open-source models to your specific domain or task with minimal computational resources
• Local AI Development: Build and test AI applications without relying on external APIs or cloud services
• Research & Experimentation: Explore reinforcement learning and advanced training techniques with built-in observability and monitoring
• Production Deployment: Export trained models in multiple formats (GGUF, safetensors) for various deployment scenarios

Who Is It For

Unsloth is designed for machine learning engineers, researchers, and developers who want to train and run open-source language models locally without expensive GPU infrastructure. It's ideal for those seeking faster iteration cycles, cost efficiency, and full control over their AI models.