### Install LLaMA Factory from source

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Clone the repository and install the package with required metrics dependencies.

```bash
git clone --depth 1 https://github.com/hiyouga/LlamaFactory.git
cd LlamaFactory
pip install -e .
pip install -r requirements/metrics.txt
```

--------------------------------

### Training Flow Example

Source: https://github.com/hiyouga/llamafactory/blob/main/CLAUDE.md

Illustrates the training invocation using a YAML configuration file.

```bash
llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
```

--------------------------------

### Install Docker (Ubuntu/Debian)

Source: https://github.com/hiyouga/llamafactory/blob/main/docker/docker-cuda/README.md

Installs Docker on Ubuntu/Debian systems. Alternatively, use the official Docker Engine installation guide.

```bash
sudo apt-get update
sudo apt-get install docker.io
```

--------------------------------

### Launch LLaMA Board GUI

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Starts the Gradio-based web interface for fine-tuning.

```bash
llamafactory-cli webui
```

--------------------------------

### Install Transformers from source

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Required step to install the latest transformers library from the main branch.

```bash
git clone -b main https://github.com/huggingface/transformers.git
cd transformers
pip install .
```

--------------------------------

### Install bitsandbytes for Ascend NPU

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Manual compilation steps for bitsandbytes on Ascend NPU environments.

```bash
# Install bitsandbytes from source
# Clone bitsandbytes repo, Ascend NPU backend is currently enabled on multi-backend-refactor branch
git clone -b multi-backend-refactor https://github.com/bitsandbytes-foundation/bitsandbytes.git
cd bitsandbytes/

# Install dependencies
pip install -r requirements-dev.txt

# Install the dependencies for the compilation tools. Note that the commands for this step may vary depending on the operating system. The following are provided for reference
apt-get install -y build-essential cmake

# Compile & install  
cmake -DCOMPUTE_BACKEND=npu -S .
make
pip install .
```

--------------------------------

### Build and run Docker containers with Docker Compose

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Commands to initialize the environment and start the container for specific hardware backends.

```bash
cd docker/docker-cuda/
docker compose up -d
docker compose exec llamafactory bash
```

```bash
cd docker/docker-npu/
docker compose up -d
docker compose exec llamafactory bash
```

```bash
cd docker/docker-rocm/
docker compose up -d
docker compose exec llamafactory bash
```

--------------------------------

### Install bitsandbytes on Windows

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Commands for installing bitsandbytes via pip or uv, including a third-party wheel option for compatibility.

```bash
pip install bitsandbytes
```

```bash
uv pip install bitsandbytes --no-deps
```

```bash
pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.2.post2-py3-none-win_amd64.whl
```

--------------------------------

### Launch Web UI ChatBox

Source: https://github.com/hiyouga/llamafactory/blob/main/examples/README.md

Start a web-based chat interface for interacting with a fine-tuned model. This is useful for a more interactive user experience.

```bash
llamafactory-cli webchat examples/inference/qwen3_lora_sft.yaml
```

--------------------------------

### Install PyTorch for Windows

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Install the CUDA-enabled version of PyTorch and verify the installation.

```bash
pip uninstall torch torchvision torchaudio
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
python -c "import torch; print(torch.cuda.is_available())"
```

--------------------------------

### Install Docker Compose (Ubuntu/Debian)

Source: https://github.com/hiyouga/llamafactory/blob/main/docker/docker-cuda/README.md

Installs Docker Compose on Ubuntu/Debian systems. The latest version can also be installed from the official Docker Compose documentation.

```bash
sudo apt-get install docker-compose
```

--------------------------------

### Run WebUI with uv

Source: https://github.com/hiyouga/llamafactory/blob/main/README.md

Launch the LLaMA Factory web interface using the uv package manager.

```bash
uv run llamafactory-cli webui
```

--------------------------------

### Install NVIDIA Container Toolkit

Source: https://github.com/hiyouga/llamafactory/blob/main/docker/docker-cuda/README.md

Installs the NVIDIA Container Toolkit, which is essential for Docker containers to access NVIDIA GPUs. This involves adding NVIDIA's repository and then installing the toolkit. Remember to restart the Docker service afterwards.

```bash
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit

sudo systemctl restart docker
```

--------------------------------

### Build Package

Source: https://github.com/hiyouga/llamafactory/blob/main/CLAUDE.md

Compiles and packages the project.

```bash
make build
```

--------------------------------

### Test CUDA Availability

Source: https://github.com/hiyouga/llamafactory/wiki/Environment

Verify that PyTorch is installed with CUDA support. Ensure CUDA is correctly configured on your system.

```python
import torch
assert torch.cuda.is_available() is True
```

--------------------------------

### FSDP+QLoRA Fine-Tuning Script

Source: https://github.com/hiyouga/llamafactory/blob/main/examples/README.md

Execute a training script that combines Fully Sharded Data Parallel (FSDP) with QLoRA for efficient fine-tuning. This is a specialized training setup.

```bash
bash examples/extras/fsdp_qlora/train.sh
```