Description
This document provides links to step-by-step instructions on how to leverage reference model docker containers to run optimized open-source Deep Learning training and inference workloads using Intel® Extension for PyTorch* and Intel® Extension for TensorFlow* on the Intel® Data Center GPU Max Series.
Base Containers
| AI Framework | Extension | Documentation |
|---|---|---|
| PyTorch | Intel® Extension for PyTorch* | Intel® Extension for PyTorch Container |
| TensorFlow | Intel® Extension for TensorFlow* | Intel® Extension for TensorFlow Container |
Optimized Workloads
The table below provides links to run each workload in a docker container. The containers were validated on a host running Linux*.
| Model | Framework | Mode | Precisions |
|---|---|---|---|
| 3D-UNet | TensorFlow | Training | BF16 |
| BERT Large | PyTorch | Inference | FP16, BF16 and FP32 |
| BERT Large | PyTorch | Training | BF16,TF32 and FP32 |
| BERT Large | TensorFlow | Training | BF16 |
| DistilBERT | PyTorch | Inference | FP16,BF16 and FP32 |
| DLRM | PyTorch | Inference | FP16 |
| DLRM | PyTorch | Training | FP32,TF32 and BF16 |
| Mask R-CNN | TensorFlow | Training | BF16 |
| ResNet50 v1.5 | PyTorch | Inference | INT8,FP16,BF16,FP32 and TF32 |
| ResNet50 v1.5 | PyTorch | Training | BF16,FP32 and TF32 |
| ResNet50 v1.5 | TensorFlow | Training | BF16 |
| RNN-T | PyTorch | Inference | FP16,BF16 and FP32 |
| RNN-T | PyTorch | Training | BF16,TF32 and FP32 |
| Stable Diffusion | PyTorch | Inference | FP16 |
Note:
- DLRM(PyTorch) inference workload is supported on older Intel® Extension for TensorFlow* v2.13 and Intel® Extension for PyTorch* 2.0.110+xpu versions.
- The other models in the list are validated on Intel® Extension for TensorFlow* v2.14 and Intel® Extension for PyTorch* 2.1.10+xpu versions.