Overview

Note: The Workload Services Framework is a benchmarking framework and is not intended to be used for the deployment of workloads in production environments. It is recommended that users consider any adjustments which may be necessary for the deployment of these workloads in a production environment including those necessary for implementing software best practices for workload scalability and security.

Introduction¶

This workload is targeting for YOLOv7 benchmarking using Pytorch framework on Intel® Xeon® processors.

The YOLO version 7 algorithm surpasses previous object detection models and YOLO versions in both speed and accuracy. It requires several times cheaper hardware than other neural networks and can be trained much faster on small datasets without any pre-trained weights. Read more at: https://viso.ai/deep-learning/yolov7-guide/

MODEL_WEIGHTS: https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7.pt
BENCHMARK_SCRIPT: https://github.com/WongKinYiu/yolov7/blob/main/test.py

The building method is mentioned in Workload Execution section.

Workload Execution¶

To execute a workload, you can refer to the executing-workload for details including cmake, ctest, terraform-options and testcase.

In the very beginning, please follow the terraform setup guide to setup terraform environment first.

Below is an execution example of terraform static test:

Create a build dir
```
cd <WSF REPO>
mkdir -p build
```

CMAKE

cd build
cmake -DBACKEND=terraform -DTERRAFORM_OPTIONS="--docker --svrinfo --owner=<your id> --intel_publish" -DTERRAFORM_SUT=static -DBENCHMARK= ..

Build the workload

cd <WSF REPO>/build/workload/YOLOv7-PyTorch-Public
make

CTEST
Show all the test cases. There are 3 diffent modes can be choosen, including latencyand throughput, and 2 case types, including gated and pkm.
```
./ctest.sh -N
```
Run specific test case(s) which shows in ./ctest.sh -N. For example, ./ctest.sh -R pkm -V is to run PKM test cases. This will use default parameters.
```
./ctest.sh -R <test case key word> -V
```
Run specific test case(s) with specified parameters which mentioned in section Parameters. For example, ./ctest.sh -R pkm --set "PRECISION=bfloat16/CORES_PER_INSTANCE=4" -V is to run PKM test case with PRECISION=bfloat16 and CORES_PER_INSTANCE=4.
```
./ctest.sh -R <test case key word> --set <specified parameters> -V
```

Parameters¶

PRECISION: Specify the model precision.
CPU: the supported precisions are float32 or bfloat16(default).
CORES_PER_INSTANCE: Specify the how many cores needed for one instance, the default value is the cores per numa node.
WARMUP_STEPS: Specify the warmup step.
CPU: the default value is 5..
STEPS: Specify the steps.
CPU: the default value is 20.
ONEDNN_VERBOSE: This parameter is only worked when HARDWARE=cpu, Specify if print the oneDNN information. you can choose 1 or 0. 1 means on, 0 means off. The default value is 0.
TORCH_TYPE: This parameter is only worked when HARDWARE=cpu, Specify the torch optimization method, you can choose of of EAGER, COMPILE-IPEX(default) or COMPILE-INDUCTOR.

Docker Image¶

This workload provides following docker image: - yolov7-pytorch-public for xeon inference benchmarking.

build docker image from scrach¶

do cmake and make to build a specific workload Please refer to cmake doc.

KPI¶

Run the list-kpi.sh script to parse the KPIs from the validation logs.

For example:

./list-kpi.sh --all logs-yolov7_pytorch_public_inference_throughput

Please refer to AI for more KPI details.

System Requirements¶

See AI Setup for system setup instructions.

Index Info¶

Name: YOLOv7-PyTorch-Public
Category: ML/DL/AI
Platform: SPR, EMR
Keywords: YOLO, CPU