openeuler/paddlepaddle Docker 镜像 - 轩辕镜像 | Docker 镜像高效稳定拉取服务

Quick reference

• The official PaddlePaddle docker image.

• Maintained by: openEuler CloudNative SIG.

• Where to get help: openEuler CloudNative SIG, openEuler.

PaddlePaddle | openEuler

Current paddlepaddle docker images are built on the openEuler. This repository is free to use and exempted from per-user rate limits.

PaddlePaddle, as the first independent R&D deep learning platform in China, has been officially open-sourced to professional communities since 2016.

Read more on PaddlePaddle Website.

Supported tags and respective Dockerfile links

The tag of each paddlepaddle docker image is consist of the version of paddlepaddle and the version of basic image. The details are as follows

Usage

In this usage, users can select the corresponding {Tag} based on their requirements.

• Pull the openeuler/paddlepaddle image from docker

  bash
  docker pull openeuler/paddlepaddle:{Tag}
  

• Run with an interactive shell

  You can also start the container with an interactive shell to use paddlepaddle.

  docker run -it --rm openeuler/paddlepaddle:{Tag} bash
  

• Introduction to PaddlePaddle with MNIST Example

  This example demonstrates how to use PaddlePaddle to build, train, evaluate, save, and load a simple LeNet-based neural network for the MNIST handwritten digit recognition task.

  • Full Example Code

  import paddle
  import numpy as np
  from paddle.vision.transforms import Normalize
  
  # 1) Load and transform MNIST dataset
  transform = Normalize(mean=[127.5], std=[127.5], data_format="CHW")
  train_dataset = paddle.vision.datasets.MNIST(mode="train", transform=transform)
  test_dataset = paddle.vision.datasets.MNIST(mode="test", transform=transform)
  
  # 2) Define the model (LeNet)
  lenet = paddle.vision.models.LeNet(num_classes=10)
  model = paddle.Model(lenet)
  
  # 3) Configure the training process
  model.prepare(
      paddle.optimizer.Adam(parameters=model.parameters()),
      paddle.nn.CrossEntropyLoss(),
      paddle.metric.Accuracy(),
  )
  
  # 4) Train the model
  model.fit(train_dataset, epochs=5, batch_size=64, verbose=1)
  
  # 5) Evaluate the model
  model.evaluate(test_dataset, batch_size=64, verbose=1)
  
  # 6) Save the trained model
  model.save("./output/mnist")
  
  # 7) Load the trained model
  model.load("output/mnist")
  
  # 8) Run inference on a single test image
  img, label = test_dataset[0]
  img_batch = np.expand_dims(img.astype("float32"), axis=0)
  out = model.predict_batch(img_batch)[0]
  pred_label = out.argmax()
  print("True label: {}, Predicted label: {}".format(label[0], pred_label))
  

  • Expected output:

  step 938/938 [==============================] - loss: 0.1575 - acc: 0.9275 - 31ms/step                            
  Epoch 2/5
  step 938/938 [==============================] - loss: 0.0990 - acc: 0.9740 - 32ms/step                            
  Epoch 3/5
  step 938/938 [==============================] - loss: 0.0196 - acc: 0.9792 - 32ms/step                           
  Epoch 4/5
  step 938/938 [==============================] - loss: 0.0052 - acc: 0.9804 - 31ms/step                           
  Epoch 5/5
  step 938/938 [==============================] - loss: 0.0253 - acc: 0.9831 - 32ms/step                               
  Eval begin...
  step 157/157 [==============================] - loss: 3.7890e-04 - acc: 0.9839 - 13ms/step                           
  Eval samples: ***
  true label: 7, pred label: 7
  

Question and answering

If you have any questions or want to use some special features, please submit an issue or a pull request on openeuler-docker-images.