DataLoader随机出现报错

[rayLemond]

trafficstars

Paddle3.0安装完后运行会报“Illegal instruction”，但机器显示支持avx指令集，看到issue区也很多同样问题没解决方案，所以降级装了paddle 2.6

当前环境： Ubuntu 22.04 CUDA 12.4 Driver Version: 550.54.14，Nvidia A800 80G nvidia-cublas-cu12 12.6.1.4 nvidia-cudnn-cu12 9.4.0.58 paddlepaddle-gpu 2.6.1.post120 paddlesci 0.0.0 python 3.12.5

问题： DataLoader会随机出现报错，同样的代码有时能正常跑完，有时报这个错

可能相关代码：

dataList = []
image = np.zeros((self.numX, self.numY))
                imageLabel = np.zeros((self.numX, self.numY))
                for i in range(bcData.shape[0]):
                    image[int(bcData[i, 0]), int(bcData[i, 1])] = bcData[i, 2]
                    imageLabel[int(bcData[i, 0]), int(bcData[i, 1])] = field_umag[i]
                    # imageLabel[int(bcData[i, 0]), int(bcData[i, 1])] = fieldData[i]
                image = np.expand_dims(image, axis=0)
                imageLabel = np.expand_dims(imageLabel, axis=0)
                data = {'image': image, 'imageLabel': imageLabel, 'index': caseName}
                dataList.append(data)

class MyDataset(paddle.io.Dataset):
    def __init__(self, dataList):
        self.dataList = dataList
        self.resizeOperator = paddle.vision.transforms.Resize(size=(256, 256))
    
    def __getitem__(self, index):
        image = self.dataList[index]['image']
        imageTensor = paddle.to_tensor(image).astype('float32')
        label = self.dataList[index]['imageLabel']
        labelTensor = paddle.to_tensor(label).astype('float32')

        return self.resizeOperator(imageTensor), self.resizeOperator(labelTensor), self.dataList[index]['index']

testDataset = MyDataset(testData.load_data(config['inChannels']))
testLoader = DataLoader(testDataset, batch_size=1, num_workers=1, shuffle=False, drop_last=False)

其他补充信息 Additional Supplementary Information

No response