Neura Market
unslothai logo

unslothai

Free

A framework that specializes in efficient fine-tuning. On its GitHub page, you can find ready-to-use fine-tuning templates for various LLMs, allowing you to easily train your own data for free on the Google Colab cloud.

FreeFree tier
Inputs: text, audioOutputs: text, code
Type
Open Source
Visit Website GitHub (38 stars)

About unslothai

Unsloth is an open-source local UI for training, running, and fine-tuning large language models (LLMs) on Windows, Linux, and macOS. It supports over 500 models including Gemma 4, Qwen3.6, DeepSeek, Kimi, GLM, and more. Unsloth offers up to 2x faster training with 70% less VRAM usage (MoE up to 12x faster), and includes features such as inference with GGUF/LoRA/safetensors support, export to GGUF and 16-bit safetensors, tool calling, code execution, an OpenAI/Anthropic-compatible API endpoint, reinforcement learning with 80% less VRAM, long-context training (3x faster, 500K+ context), multi-GPU training, data recipes for automatic dataset creation from PDFs/CSVs/DOCX, and MCP control for managing models and training. Unsloth Studio (beta) provides a visual interface for running text, audio, embedding, and vision models locally.

Key Features

Unsloth Studio (Beta) for running and training text, audio, embedding, and vision models locally
Inference with search, download, and run support for GGUF, LoRA adapters, and safetensors
Export models to GGUF and 16-bit safetensors
Self-healing tool calling and web search
Code execution in sandbox environments
OpenAI/Anthropic-compatible API endpoint for local models (/v1/chat/completions, etc.)
Train and RL 500+ models up to 2x faster with 70% less VRAM (MoE up to 12x faster)
Reinforcement learning with 80% less VRAM for GRPO, FP8, and vision RL
Long-context training: 3x faster, 30% less VRAM, and 500K+ context
Data Recipes for auto-creating datasets from PDF, CSV, DOCX, etc.

Pros & Cons

Pros
  • Up to 2x faster training with 70% lower VRAM usage compared to standard methods
  • Supports over 500 models including latest releases like Gemma 4, Qwen3.6, DeepSeek
  • Runs locally on Windows, Linux, macOS, and AMD GPUs
  • Includes a built-in inference server with OpenAI/Anthropic-compatible API
  • Supports export to GGUF and safetensors for easy deployment
  • Data Recipes automate dataset creation from common document formats
  • Open-source and free to use with active community on GitHub
Cons
  • Requires local GPU hardware with sufficient VRAM for optimal performance
  • Unsloth Studio is in beta, so some features may be unstable or incomplete
  • Setup may require technical expertise (command-line installation)
  • Documentation is primarily on GitHub and may lack beginner-friendly guides

Best For

Fine-tuning and RL on large language models locallyRunning inference on text, audio, embedding, and vision modelsDeploying local LLMs as API endpoints for agents like Claude Code and CodexComparing models side by side with the same promptsCreating training datasets from PDFs, CSVs, DOCX using Data RecipesLong-context training for documents up to 500K+ tokensExporting trained models to GGUF or safetensors for use in other applications

FAQ

What is Unsloth?
Unsloth is an open-source local UI for training and running large language models. It supports 500+ models and offers up to 2x faster training with 70% less VRAM.
How do I install Unsloth?
On macOS, Linux, or WSL, run: curl -fsSL https://unsloth.ai/install.sh | sh. On Windows, run: irm https://unsloth.ai/install.ps1 | iex. Docker and source builds are also available.
What models are supported?
Unsloth supports over 500 models including Gemma 4, Qwen3.6, DeepSeek, Kimi, GLM, Llama 4, Mistral, Phi-4, and many more from Hugging Face.
Can I export trained models?
Yes, Unsloth allows exporting models to GGUF, 16-bit safetensors, and other formats for use in other applications.
Does Unsloth support fine-tuning?
Yes, Unsloth supports fine-tuning, including LoRA/QLoRA, full fine-tuning, reinforcement learning, pretraining, and long-context training up to 500K+ tokens.
Can I use Unsloth with agents like Claude Code?
Yes, Unsloth Start connects local models to agents such as Claude Code, Codex, and Hermes via a single command (e.g., unsloth start claude).