From c60636695c8ea4fd71989ae4a0195302396a1a9a Mon Sep 17 00:00:00 2001 From: Michael Han <107991372+shimmyshimmer@users.noreply.github.com> Date: Tue, 17 Mar 2026 07:53:50 -0700 Subject: [PATCH] Unsloth Studio.md --- README.md | 453 +++++++++--------------------------------------------- 1 file changed, 76 insertions(+), 377 deletions(-) diff --git a/README.md b/README.md index 1314cb1c59..429311baf3 100644 --- a/README.md +++ b/README.md @@ -1,24 +1,83 @@ -
+
-
-
++ Features • + Quickstart • + Notebooks • + Documentation • + Discord +
+
-
+Unsloth Studio lets you run and train models for text, [audio](https://unsloth.ai/docs/basics/text-to-speech-tts-fine-tuning), [embedding](https://unsloth.ai/docs/new/embedding-finetuning), [vision](https://unsloth.ai/docs/basics/vision-fine-tuning) and more. Available on Windows, Linux and macOS.
+## ⭐ Features
+Unsloth provides several key features for both inference and training:
+### Inference
+* **Search + download + run models** including GGUF, LoRA adapters, safetensors
+* **Export models**: [Save or export](https://unsloth.ai/docs/new/studio/export) models to GGUF, 16-bit safetensors and other formats.
+* **Tool calling**: Support for [self-healing tool calling](https://unsloth.ai/docs/new/studio/chat#auto-healing-tool-calling) and web search
+* **[Code execution](https://unsloth.ai/docs/new/studio/chat#code-execution)**: lets LLMs run code, data and verify results so answers are more accurate.
+* [Auto-tune inference parameters](https://unsloth.ai/docs/new/studio/chat#auto-parameter-tuning) and customize chat templates.
+* Upload images, audio, PDFs, code, DOCX and more file types to chat with.
+### Training
+* Train **500+ models** up to **2x faster** with up to **70% less VRAM**, with no accuracy loss.
+* Supports full fine-tuning, pretraining, 4-bit, 16-bit and, FP8 training.
+* **Observability**: Monitor training live, track loss and GPU usage and customize graphs.
+* **Data Recipes**: [Auto-create datasets](https://unsloth.ai/docs/new/studio/data-recipe) from **PDF, CSV, DOCX** etc. Edit data in a visual-node workflow.
+* **Reinforcement Learning**: The most efficient [RL](https://unsloth.ai/docs/get-started/reinforcement-learning-rl-guide) library, using **80% less VRAM** for GRPO, [FP8](https://unsloth.ai/docs/get-started/reinforcement-learning-rl-guide/fp8-reinforcement-learning) etc.
+* [Multi-GPU](https://unsloth.ai/docs/basics/multi-gpu-training-with-unsloth) training is supported, with major improvements coming soon.
-
-
-## 💾 Install Unsloth
-You can also see our docs for more detailed installation and updating instructions [here](https://unsloth.ai/docs/get-started/install).
-
-Unsloth supports Python 3.13 or lower.
-
-### Pip Installation
-**Install with pip (recommended) for Linux devices:**
-```
-pip install unsloth
-```
-**To update Unsloth:**
-```
-pip install --upgrade --force-reinstall --no-cache-dir unsloth unsloth_zoo
-```
-See [here](#advanced-pip-installation) for advanced pip install instructions.
-### Windows Installation
-For this method, we will be utilizing Anaconda. You can view the [full guide with screenshots here](https://unsloth.ai/docs/get-started/install/windows-installation).
-1. **Install Miniconda (or Anaconda):** Miniconda is recommended. Install [Miniconda](https://www.anaconda.com/docs/getting-started/miniconda/install) or [Anaconda](https://www.anaconda.com/download), then open Anaconda PowerShell Prompt to continue.
-
-2. **Create a Conda Environment:** Create and activate a fresh Python 3.12 environment for Unsloth.
-
- ```bash
- conda create --name unsloth_env python==3.12 -y
- conda activate unsloth_env
- ```
-
-3. **Check Your GPU and CUDA Version:** Run `nvidia-smi` to confirm that your NVIDIA GPU is detected and note the CUDA version shown in the output. If `nvidia-smi` does not work, reinstall the latest [NVIDIA drivers](https://www.nvidia.com/en-us/drivers/).
-
-4. **Install PyTorch:** Install the Windows pip build of PyTorch that matches your CUDA version. Use [Install PyTorch](https://pytorch.org/get-started/locally/) to select the correct command for your system, then verify that PyTorch can see your GPU.
-
- ```python
- import torch
- print(torch.cuda.is_available())
- A = torch.ones((10, 10), device="cuda")
- B = torch.ones((10, 10), device="cuda")
- A @ B
- ```
-
-5. **Install Unsloth:** Only install Unsloth after PyTorch is working correctly.
-
- ```bash
- pip install unsloth
- ```
-
-#### Advanced/Troubleshooting
-For **advanced installation instructions** or if you see weird errors during installations:
-
-First try using an isolated environment via then `pip install unsloth`
-```bash
-python -m venv unsloth
-source unsloth/bin/activate
-pip install unsloth
-```
-
-1. Install `torch` and `triton`. Go to https://pytorch.org to install it. For example `pip install torch torchvision torchaudio triton`
-2. Confirm if CUDA is installed correctly. Try `nvcc`. If that fails, you need to install `cudatoolkit` or CUDA drivers.
-3. Install `xformers` manually via:
- ```bash
- pip install ninja
- pip install -v --no-build-isolation -U git+https://github.com/facebookresearch/xformers.git@main#egg=xformers
- ```
- Check if `xformers` succeeded with `python -m xformers.info` Go to https://github.com/facebookresearch/xformers. Another option is to install `flash-attn` for Ampere GPUs and ignore `xformers`
-
-4. For GRPO runs, you can try installing `vllm` and seeing if `pip install vllm` succeeds.
-5. Double check that your versions of Python, CUDA, CUDNN, `torch`, `triton`, and `xformers` are compatible with one another. The [PyTorch Compatibility Matrix](https://github.com/pytorch/pytorch/blob/main/RELEASE.md#release-compatibility-matrix) may be useful.
-6. Finally, install `bitsandbytes` and check it with `python -m bitsandbytes`
-
-### Conda Installation (Optional)
-`⚠️Only use Conda if you have it. If not, use Pip`. We support `python=3.10,3.11,3.12,3.13`.
-```bash
-conda create --name unsloth_env python==3.12 -y
-conda activate unsloth_env
-```
-Use `nvidia-smi` to get the correct CUDA version like 13.0 which becomes `cu130`
-```bash
-pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130
-pip3 install unsloth
-```
-
### Thank You to
-- The [llama.cpp library](https://github.com/ggml-org/llama.cpp) that lets users save models with Unsloth
+- The [llama.cpp library](https://github.com/ggml-org/llama.cpp) that lets users run and save models with Unsloth
- The Hugging Face team and their libraries: [transformers](https://github.com/huggingface/transformers) and [TRL](https://github.com/huggingface/trl)
- The Pytorch and [Torch AO](https://github.com/unslothai/unsloth/pull/3391) team for their contributions
- And of course for every single person who has contributed or has used Unsloth!