smaller
bigger
print

Home |
French |
Search |

Institute for Development and Resources in Intensive Scientific Computing

Navigation du site

IDRIS

Overview
Staff
Supercomputers
Training courses
Contact us
Location Map

Resource Management

Scientific Management
Requesting resource allocations
Account Management
Registration of user machines
Forms
Extranet

For Users

Getting started
Machine availability
HPE SGI 8600 (Jean Zay)
Atomistic simulation
Artificial intelligence
FAQ
User support
Advanced support
Users Committee

News

PRACE Projects

Table des matières

Artificial Intelligence research on Jean Zay

Artificial Intelligence research on Jean Zay

This page provides useful information for researchers in artificial intelligence who are working on Jean Zay. The following is an extract from the complete documentation available on the page Jean Zay.

Tools available on Jean Zay

Getting started

Getting Started on Jean Zay for AI users

Requesting resource hours at IDRIS

Computing environment

Software for artificial intelligence
Python
Personal Python environment
Access to JupyterHub (JupyterLab, TensorBoard,...)
Datasets available in $DSDIR storage space
Database management

Job execution on GPU partitions with Slurm

GPU Slurm partitions
Batch job commands
Interactive execution
Single-GPU code in a batch job
Collection of similar jobs: Job Array
SSH connection to a compute node
Python scripts for automated execution of GPU jobs(idr_pytools)

Visualisation and post-processing tools

TensorBoard visualisation tool for TensorFlow and PyTorch
TensorFlow and PyTorch profiling tools

Deep neural network learning

Optimisation of GPU acceleration

Using mixed precision to optimise model learning
Accelerating calculations with the use of Tensor Cores

Techniques of learning distribution on Jean Zay

Multi-GPU and multi-node distribution for training a TensorFlow or PyTorch model
Database loading for the distributed learning of a model
SyncBN layer for data parallelism

Benchmarks and best practices on Jean Zay

Performance tests

Performance tests of a Deep Learning training with and without Singularity containerisation

Other applications in data science, Machine Learning, Big Data

RAPIDS: Accelerated data science on GPUs

CNRS.fr
Extranet
FAQ
Archives
Contacts
Location Map
Legal Notices