Unsloth

Unsloth is an open-source tool and no-code web UI (Unsloth Studio) for developers, researchers, and teams to train, fine-tune, and run open models locally on Mac and Windows.

Key capabilities include:

• Offline local runtime for GGUF and Safetensors models, plus tool-calling, web search, and an OpenAI-compatible API for app integration
• No-code training workflows, including auto dataset creation from PDF, CSV, and JSON, with real-time observability during training
• Broad model support (500+ models) across text, vision, audio, and embeddings, including families such as Llama, Mistral, Gemma, and Phi-3
• Optimized fine-tuning kernels for LoRA/QLoRA and 4-bit/16-bit modes (plus FP8, FFT, PT), designed to reduce VRAM usage and accelerate training on NVIDIA GPUs

Pricing: Freemium. Free open-source version; Pro and Enterprise plans are available via contact.

Notable for combining a local, offline Studio experience with export options (GGUF/Safetensors) and performance-focused kernels that can run on local machines, Google Colab, or scale toward multi-GPU and multi-node setups.