Model Zoo for Intel® Architecture v2.7.0

Supported Frameworks

• TensorFlow v2.8.0
• PyTorch v1.11.0 and IPEX v1.11.0

New models

• N/A

New features

• Transfer Learning notebooks for NLP and Computer Vision: https://github.com/IntelAI/models/tree/v2.7.0/docs/notebooks/transfer_learning
• Consolidate TensorFlow and PyTorch benchmark tables based on the use case: https://github.com/IntelAI/models/tree/v2.7.0#use-cases
• Added links for required dataset for each use case: https://github.com/IntelAI/models/tree/v2.7.0/benchmarks
• Initial support for running several models on Windows platform: https://github.com/IntelAI/models/blob/master/docs/general/tensorflow/Windows.md
• Experimental support for running models on CentOS 8 Stream, Red Hat 8 and SLES 15

Bug fixes:

• This release contains many bug fixes to the previous versions. Please see the commit history here: https://github.com/IntelAI/models/commits/v2.7.0

Supported Configurations

Intel Model Zoo 2.7.0 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.8, 3.9
• Docker Server v19+
• Docker Client v18+

Model Zoo for Intel® Architecture v2.6.1

Features and bug fixes

• Update ImageNet dataset preprocessing instructions here: datasets/imagenet

Supported Configurations

Intel Model Zoo 2.6.1 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.8, 3.9
• Docker Server v19+
• Docker Client v18+

Model Zoo for Intel® Architecture v2.6.0

TensorFlow Framework

• Support for TensorFlow v2.7.0

New TensorFlow models

• N/A

Other features and bug fixes for TensorFlow models

• Updates to only use docker --privileged when required and check --cpuset
• • Except for BERT Large and Wide and Deep models
• Updated the ImageNet download link
• Fix platform_util.py for systems with only one socket or subset of cores within a socket
• Replace USE_DAAL4PY_SKLEARN env var with patch_sklearn
• Add error handling for when a frozen graph isn't passed for BERT large FP32 inference*

PyTorch Framework

• Support for PyTorch v1.10.0 and IPEX v1.10.0

New PyTorch models

• GoogLeNet Inference(FP32, BFloat16**)
• Inception v3 Inference(FP32, BFloat16**)
• MNASNet 0.5 Inference(FP32, BFloat16**)
• MNASNet 1.0 Inference(FP32, BFloat16**)
• ResNet 50 Inference(Int8)
• ResNet 50 Training(FP32, BFloat16**)
• ResNet 101 Inference(FP32, BFloat16**)
• ResNet 152 Inference(FP32, BFloat16**)
• ResNext 32x4d Inference(FP32, BFloat16**)
• ResNext 32x16d Inference(FP32, Int8, BFloat16**)
• VGG-11 Inference(FP32, BFloat16**)
• VGG-11 with batch normalization Inference(FP32, BFloat16**)
• Wide ResNet-50-2 Inference(FP32, BFloat16**)
• Wide ResNet-101-2 Inference(FP32, BFloat16**)
• BERT base Inference(FP32, BFloat16**)
• BERT large Inference(FP32, Int8, BFloat16**)
• BERT large Training(FP32, BFloat16**)
• DistilBERT base Inference(FP32, BFloat16**)
• RNN-T Inference(FP32, BFloat16**)
• RNN-T Training(FP32, BFloat16**)
• RoBERTa base Inference(FP32, BFloat16**)
• Faster R-CNN ResNet50 FPN Inference(FP32
• Mask R-CNN Inference(FP32, BFloat16**)
• Mask R-CNN Training(FP32, BFloat16**)
• Mask R-CNN ResNet50 FPN Inference(FP32)
• RetinaNet ResNet-50 FPN Inference(FP32)
• SSD-ResNet34 Inference(FP32, Int8, BFloat16**)
• SSD-ResNet34 Training(FP32, BFloat16**)
• DLRM Inference(FP32, Int8, BFloat16**)
• DLRM Training(FP32)

Other features and bug fixes for PyTorch models

• DLRM and ResNet 50 documentation updates

Supported Configurations

Intel Model Zoo 2.6.0 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.8, 3.9
• Docker Server v19+
• Docker Client v18+

Intel Model Zoo v2.5.0

New Functionality

New Models

• ML-Perf Transformer-LT Training (FP32 and BFloat16)
• ML-Perf Transformer-LT Inference (FP32, BFloat16 and INT8)
• ML-Perf 3D-Unet Inference (FP32, BFloat16 and INT8)
• DIEN Training (FP32)
• DIEN Inference (FP32 and BFloat16)

Other features and bug fixes

• Added IPython Notebook with BERT classifier fine tuning using IMDb
• Documentation for creating an LPOT Container with Intel® Optimizations for TensorFlow
• Advanced documentation for wide deep large ds fp32 training
• Increase Unit testing coverage

DL Frameworks (TensorFlow)

• Support for TensorFlow v2.6.0 and TensorFlow Serving v2.6.0

DL Frameworks (PyTorch)

• Support for PyTorch v1.9.0 and IPEX v1.9.0

Supported Configurations

Intel Model Zoo 2.5.0 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.8
• Docker Server v19+
• Docker Client v18+

v2.4.0

New Functionality

• Added links to Intel oneContainer Portal
• Added documentation for running most workflows inside Intel® oneAPI AI Analytics Toolkit
• Experimental support for running workflows onCentOS 8

DL Frameworks (TensorFlow)

• Support for TensorFlow v2.5.0 and TensorFlow Serving v2.5.1

Supported Configurations

Intel Model Zoo 2.4 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.8
• Docker Server v19+
• Docker Client v18+

v2.3.0

New Functionality

Two new TF workload containers and model packages that are available on the Intel® oneContainer Portal:

• 3D U-Net FP32 Inference
• SSD ResNet34 BFloat16 Training

One new PyTorch workload containers and model packages that are available on the Intel® oneContainer Portal:

• DLRM BFloat16 Training

DL Frameworks (TensorFlow)

TensorFlow models in the 2.3 release are validated on the following TensorFlow versions:

• Intel Optimizations for TensorFlow v2.4.0 or v1.15.2 (select models)
• Intel Optimizations for TensorFlow serving v2.3

DL Frameworks (PyTorch)

PyTorch models in the 2.3 release are validated on the following PyTorch version:

• PyTorch v1.5.0-rc3

Supported Configurations

Intel Model Zoo 2.3 is validated on the following environment:

• Ubuntu 20.04 LTS
• Python 3.6
• Docker Server v19+
• Docker Client v18+

Intel Model Zoo 2.2.0 release

New Functionality

Ten new TF workload containers and model packages that are available on the Intel® oneContainer Portal:

• InceptionV4 FP32 Inference
• InceptionV4 Int8 Inference
• ResNet101 FP32 Inference
• ResNet101 Int8 Inference
• Transformer-LT MLPerf BFloat16 Training
• GNMT MLPerf FP32 Inference
• DenseNet169 FP32 Inference
• ResNet50v1.5 BFloat16 Training
• Wide & Deep Large Dataset FP32 Inference
• Wide & Deep Large Dataset Int8 Inference

Two new PyTorch workload containers and model packages that are available on the Intel® oneContainer Portal:

• ResNet50 FP32 Inference
• ResNet50 BFloat16 Inference

Three new TF Kubernetes packages that are available on the Intel® oneContainer Portal:

• BERT Large FP32 Training
• Wide & Deep Large Dataset FP32 Training
• RFCN FP32 Inference

A new Helm chart to deploy TensorFlow Serving on a K8s cluster

DL Frameworks (TensorFlow)

TensorFlow models in the 2.2 release are validated on the following TensorFlow versions:

• Intel Optimizations for TensorFlow v2.3.0 or v1.15.2 (select models)
• Intel Optimizations for TensorFlow serving v2.3

DL Frameworks (PyTorch)

PyTorch models in the 2.2 release are validated on the following PyTorch version:

• PyTorch v1.5.0-rc3

Supported Configurations

Intel Model Zoo 2.2 is validated on the following environment:

• Ubuntu 18.04 LTS
• Python 3.6
• Docker Server v19+
• Docker Client v18+

Intel Model Zoo v2.1.1

New Functionality

One new TensorFlow workload containers that are available on the Intel® oneContainer Portal:

• COCO Validation Dataset Preprocessing

6 new TensorFlow workload containers that are available on the Intel® oneContainer Portal:

• TensorFlow Performance Comparison Jupyter Notebook

• Intel PyTorch Extension

• Intel Optimizations for TensorFlow with Scikit-learn and oneDAL

• Scikit-learn with oneDAL

• XGBoost with oneDAL

• Intel Optimized Analytics Package(OAP) for Spark Platform

One new TensorFlow K8s workflow package that are available on the Intel® oneContainer Portal:

• ResNet50 v1.5 FP32 Training

DL Frameworks (TensorFlow)

TensorFlow models in this release are validated on the following TensorFlow versions:

• Intel Optimizations for TensorFlow v2.3.0 or v1.15.2 (select models)

• Intel Optimizations for TensorFlow Serving v2.3.0

DL Frameworks (PyTorch)

• PyTorch models in this release are validated on the following PyTorch version:

• PyTorch v1.5.0-rc3

Supported Configurations

• Intel Model Zoo 2.0 is validated on the following environment:

• Ubuntu 18.04 LTS

• Python 3.6

• Docker Server v19+

• Docker Client v18+