Reference Models for Voyager

This repository contains the necessary files and links to run a collection of models on Voyager at the San Diego Supercomputer Center (SDSC). The majority of those models are the ones supported by Intel Habana link.

The last column on the list (Verified) indicates the last version of Synapse AI the model was tested on Voyager.

The list of models is being improved. Your feedback is greatly appreciated. Feel free to open an issue or contact me (Javier Hernandez-Nicolau) at javierhn *at* ucsd.edu.

Model list

Simple examples

• 'Hello World!': simplest example to run on Voyager.
• Port your Pytorch application. Tutorial to port a Pytorch application.
• (OLD) Port your Pytorch application. A porting based on the Lazy mode.
• Running Jupyter Notebook. A tutorial on running Jupyter Notebook on Voyager.
• MPIJob PT Learn how to run a MNIST Pytorch model in multiple nodes.
• Parallelize your application with DDP. Tutorial to parallelize a working Pytorch app.
• Tutorial: Huggingface Diffusers Learn ho to run Stable Diffusion Inference on HPUs.

Computer Vision

Models	Framework	Multi-node	Verified
ResNet50, ResNeXt101	Pytorch	Yes	1.21.4
ResNet50 (Pytorch Lightning)	Pytorch Lightning		1.15.1
UNet2D,UNet3D	Pytorch Lightning		1.21.4
YOLOX	Pytorch	Yes	1.21.4
SSD	Pytorch		1.21.4
Vision Transformer	Pytorch	Yes	1.21.4
UNet-2D	TensorFlow	Yes	1.15.1
ResNet50 (Keras)	TensorFlow	Yes	1.15.1
ResNeXt101	TensorFlow	Yes	1.15.1

Natural Language processing

Models	Framework	Multi-node	Verified
BERT	Pytorch	Yes	1.21.4
BART (fine-tuning,simpletransformers)	Pytorch	Yes	1.15.1
Hugginface BLOOM (inference)	Pytorch		1.15.1
BERT 1.5B and 5B with Deepspeed	Pytorch	Yes	1.15.1
LLaMA (Megatron-DeepSpeed)	Pytorch	Yes	1.13
BERT	TensorFlow		1.13

Generative models

Models	Framework	Multi-node	Verified
(Not supported now) Stable-Diffusion	Pytorch		1.11
Stable-Diffusion 2.1 (Inference, Huggingface)	Pytorch		1.15.1