AI Help

From UFRC
Revision as of 14:15, 17 June 2021 by Ericeric (talk | contribs)
Jump to navigation Jump to search

New User's Guide

For new users on HiPerGator, please read Getting Started to get yourself familiar with HiPerGator system.

For beginners, we have a series of pre-recorded training videos available on a variety of topics. It is highly recommended that you watch these videos, such as Introduction to HiPerGator, SLURM job submission, and using Jupyter Hub and Jupyter Notebooks, etc.

AI Software

A comprehensive software stack for AI research is available on HiPerGator for both CPU and GPU accelerated applications.

AI Frameworks

AI frameworks provide building blocks for designing and training machine learning and deep learning models. The following AI frameworks are available on HiPerGator.

PyTorch

PyTorch is a deep learning framework developed by Facebook AI Research Lab and has interfaces for Python, Java, and C++, but is most commonly used with Python. It supports training on both GPU and CPUs, as well as distributed training and multi-GPU models. See our PyTorch quickstart page for help getting started using PyTorch on HiPerGator.

Tensorflow/Keras

TensorFlow is an open-source AI framework/platform developed by the Google Brain team. Keras is an open-source neural network library which runs on top of TensorFlow. With TensorFlow 2.0, the Keras API has been integrated in TensorFlow's core library and serves as a high-level Python interface for TensorFlow. TensorFlow supports both GPU and CPUs, as well as multi-GPU and distributed training. APIs are available for Python, Java, Go and C++. See our TensorFlow quickstart page for help getting started using TensorFlow on HiPerGator.

Sci-kit Learn

Sci-kit learn is a Python library for machine learning and statistical modeling. It is available in many of the Python modules on HiPerGator.

MATLAB

Matlab provides convenient toolboxes for machine learning, deep learning, computer vision and automatic driving, which are supported on both CPUs and GPUs.

NVIDIA AI Software

Nvidia provides comprehensive, GPU-accelerated software libraries, toolkits, frameworks and packages for big-data and AI applications. Many of the libraries are included in the CUDA installation on HiPerGator, such as cuDNN. The following domain specific CUDA enabled tools are available on HiPerGator:

Megatron-LM

Megatron-LM can train several architectures of language models, including an GPT, T5, and an improved BERT. Megatron also recently added a transformer-based image classification architecture.

NeMo

Nvidia NeMo is an open-source Python, GPU-accelerated toolkit for conversational AI, including speech recognition (ASR), natural language processing (NLP) and text to speech (TTS) applications. NeMo is available via a container in the apps folder.

Clara Parabricks

Nvidia Clara Parabricks ([1]) is a computation framework for genomics applications. It builds GPU accelerated libraries, pipelines, and reference AI workflows for genomics research. We have a license for this software through 2021.

NLP Environments

There are five environments primarily oriented to NLP. Two of the environments are available via module load or as kernels in Jupyter notebooks - nlp and bertopic. nlp has Pytorch 1.8.1, torchtext, Pytorch Lightning, spacy, nltk, sklearn, and rapids. Bertopic has Tensorflow 2.4, Torch 1.7, transformers, sentence transformers, sentence-piece, nltk, and sklearn. We created a singularity container from Nvidia's Pytorch container, this contains all the dependencies required for Megatron-LM. That and a container for NeMo can be found in /apps/nvidia/containers. In addition, there is spark-nlp for those using Spark clusters, module load spark and tensorflow/2.4.1.