Skip to content

Huawei Ascend NPU Quick Start Guide

This section will guide you through the process of quickly using the Huawei Ascend NPU for inference tasks.

Prerequisites

  1. Install the Ascend NPU driver.

    Make sure you have installed the Ascend NPU driver and can run the npu-smi info command, which should return NPU information indicating that the driver and firmware are ready.

    Ascend Information

  2. Install the Ascend NPU Device Plugin.

    The NPU Device Plugin is installed by default in the kube-system namespace. It is a DaemonSet workload that can be verified by running the command kubectl get pod -n kube-system | grep ascend , which should output the following:

    Ascend Device Plugin

    If the driver and Device Plugin are not installed, refer to the official Ascend documentation for installation instructions:

  3. Download the Ascend code repository.

    Run the following command to download the Ascend sample code repository and remember the location where the code is stored, as it will be needed later.

    git clone https://gitee.com/ascend/samples.git

Quick Start

This guide uses the AscentCL Image Classification Application example from the Ascend sample repository.

Prepare the base image

This example uses the Ascent-pytorch base image, which can be obtained from the Ascend Image Repository.

Edit YAML

apiVersion: batch/v1
kind: Job
metadata:
  name: resnetinfer1-1-1usoc
spec:
  template:
    spec:
      containers:
        - image: ascendhub.huawei.com/public-ascendhub/ascend-pytorch:23.0.RC2-ubuntu18.04 # Inference image name
          imagePullPolicy: IfNotPresent
          name: resnet50infer
          securityContext:
            runAsUser: 0
          command:
            - "/bin/bash"
            - "-c"
            - |
              source /usr/local/Ascend/ascend-toolkit/set_env.sh &&
              TEMP_DIR=/root/samples_copy_$(date '+%Y%m%d_%H%M%S_%N') &&
              cp -r /root/samples "$TEMP_DIR" &&
              cd "$TEMP_DIR"/inference/modelInference/sampleResnetQuickStart/python/model &&
              wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/003_Atc_Models/resnet50/resnet50.onnx &&
              atc --model=resnet50.onnx --framework=5 --output=resnet50 --input_shape="actual_input_1:1,3,224,224"  --soc_version=Ascend910 &&
              cd ../data &&
              wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/models/aclsample/dog1_1024_683.jpg &&
              cd ../scripts &&
              bash sample_run.sh
          resources:
            requests:
              huawei.com/Ascend910: 1 # Number of the Ascend 910 Processors.
            limits:
              huawei.com/Ascend910: 1 # The value should be the same as that of requests .
          volumeMounts:
            - name: hiai-driver
              mountPath: /usr/local/Ascend/driver
              readOnly: true
            - name: slog
              mountPath: /var/log/npu/conf/slog/slog.conf
            - name: localtime #The container time must be the same as the host time.
              mountPath: /etc/localtime
            - name: dmp
              mountPath: /var/dmp_daemon
            - name: slogd
              mountPath: /var/slogd
            - name: hbasic
              mountPath: /etc/hdcBasic.cfg
            - name: sys-version
              mountPath: /etc/sys_version.conf
            - name: aicpu
              mountPath: /usr/lib64/aicpu_kernels
            - name: tfso
              mountPath: /usr/lib64/libtensorflow.so
            - name: sample-path
              mountPath: /root/samples
      volumes:
        - name: hiai-driver
          hostPath:
            path: /usr/local/Ascend/driver
        - name: slog
          hostPath:
            path: /var/log/npu/conf/slog/slog.conf
        - name: localtime
          hostPath:
            path: /etc/localtime
        - name: dmp
          hostPath:
            path: /var/dmp_daemon
        - name: slogd
          hostPath:
            path: /var/slogd
        - name: hbasic
          hostPath:
            path: /etc/hdcBasic.cfg
        - name: sys-version
          hostPath:
            path: /etc/sys_version.conf
        - name: aicpu
          hostPath:
            path: /usr/lib64/aicpu_kernels
        - name: tfso
          hostPath:
            path: /usr/lib64/libtensorflow.so
        - name: sample-path
          hostPath:
            path: /root/samples
      restartPolicy: OnFailure

There are some fields in the above YAML that need to be modified according to your specific situation:

  1. atc ... --soc_version=Ascend910 is using Ascend910 as an example. Please replace it with the actual model you are using. You can use the npu-smi info command to check the model of your card and add the Ascend prefix accordingly.
  2. samples-path should be replaced with the actual path in your system.
  3. resources should be adjusted based on your specific requirements.

Deploy Job and View Results

Use the following command to create a Job:

kubectl apply -f ascend-demo.yaml

Check the Pod status:

Ascend Pod Status

Once the Pod is successfully running, view the log results. The key information displayed on the screen is as shown in the following image. The label in the prompt represents the category label, conf represents the maximum confidence of that category, and class represents the corresponding category. These values may vary depending on the version and environment, so refer to your actual situation.

Ascend demo execution result

Result is as follows:

Ascend demo inference result image

Comments