Skip to content

Latest commit

 

History

History
108 lines (83 loc) · 4.75 KB

INSTALL.md

File metadata and controls

108 lines (83 loc) · 4.75 KB
Raw

h2oGPT Installation Help

The following sections describe how to get a working Python environment on a Linux system.

Install for A100+

E.g. for Ubuntu 20.04, install driver if you haven't already done so:

sudo apt-get update
sudo apt-get -y install nvidia-headless-535-server nvidia-fabricmanager-535 nvidia-utils-535-server
# sudo apt-get -y install nvidia-headless-no-dkms-535-servers

Note that if you run the preceding commands, you don't need to use the NVIDIA developer downloads in the following sections.

Install CUDA Toolkit

If happy with above drivers, then just get run local file for CUDA 11.8:

wget wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
sudo sh cuda_11.8.0_520.61.05_linux.run

only choose to install toolkit and do not replace existing /usr/local/cuda link if you already have one.

If instead, you want full deb CUDA install cuda coolkit. Pick deb local, e.g. for Ubuntu:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda-repo-ubuntu2004-12-1-local_12.1.0-530.30.02-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu2004-12-1-local_12.1.0-530.30.02-1_amd64.deb
sudo cp /var/cuda-repo-ubuntu2004-12-1-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda

Then set the system up to use the freshly installed CUDA location:

echo "export LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/cuda/lib64/" >> ~/.bashrc
echo "export CUDA_HOME=/usr/local/cuda" >> ~/.bashrc
echo "export PATH=\$PATH:/usr/local/cuda/bin/" >> ~/.bashrc
source ~/.bashrc

Then reboot the machine, to get everything sync'ed up on restart.

sudo reboot

Compile bitsandbytes

For fast 4-bit and 8-bit training, you need to use bitsandbytes. Note that compiling bitsandbytes is only required if you have a different CUDA version from the ones built into the bitsandbytes PyPI package, which includes CUDA 11.0, 11.1, 11.2, 11.3, 11.4, 11.5, 11.6, 11.7, 11.8, 12.0, and 12.1. In the following example, bitsandbytes is compiled for CUDA 12.1:

git clone http://github.com/TimDettmers/bitsandbytes.git
cd bitsandbytes
git checkout 7c651012fce87881bb4e194a26af25790cadea4f
CUDA_VERSION=121 make cuda12x
CUDA_VERSION=121 python setup.py install
cd ..

Install NVIDIA GPU Manager on systems with multiple A100 or H100 GPUs

To install NVIDIA GPU Manager, run the following:

sudo apt-key del 7fa2af80
distribution=$(. /etc/os-release;echo $ID$VERSION_ID | sed -e 's/\.//g')
wget https://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64/cuda-keyring_1.0-1_all.deb
sudo dpkg -i cuda-keyring_1.0-1_all.deb
sudo apt-get update
sudo apt-get install -y datacenter-gpu-manager
# if use 535 drivers, then use 535 below
sudo apt-get install -y libnvidia-nscq-535
sudo systemctl --now enable nvidia-dcgm
dcgmi discovery -l

For more information, see the official GPU Manager user guide.

Install and run NVIDIA Fabric Manager on systems with multiple A100 or H100 GPUs

To install the CUDA drivers for NVIDIA Fabric Manager, run the following:

sudo apt-get install -y cuda-drivers-fabricmanager

Once you've installed Fabric Manager and rebooted your system, run the following to start the NVIDIA Fabric Manager service:

sudo systemctl --now enable nvidia-dcgm
dcgmi discovery -l
sudo systemctl start nvidia-fabricmanager
sudo systemctl status nvidia-fabricmanager

For more information, see the official Fabric Manager user guide.

Optional: Use TensorBoard to inspect training

You can use TensorBoard to inspect the training process. To launch TensorBoard and instruct it to read event files from the runs/ directory, use the following command:

tensorboard --logdir=runs/

For more information, see TensorBoard usage.