MXNet maven GPU error on Windows OS.

[iamuspace]

I want to use DJL engine maven with Cuda on Windows PC.

My env is
Windows 10, AMD cpu 64bit, Zulu 64bit JVM, Nvidia Titan X GPU, maven

I run ok with MXNet cpu maven.

but, runtime error with MXNet GPU engine maven :

[11:59:02] C:\Users\Administrator\kimbergz\b4\src\imperative./imperative_utils.h:93: GPU support is disabled. Compile MXNet with USE_CUDA=1 to enable GPU support.
Exception in thread "main" ai.djl.engine.EngineException: MXNet engine call failed: MXNetError: GPU is not enabled
Stack trace:
File "..\src\resource.cc", line 167

at ai.djl.mxnet.jna.JnaUtils.checkCall(JnaUtils.java:1942)
at ai.djl.mxnet.jna.JnaUtils.imperativeInvoke(JnaUtils.java:521)
at ai.djl.mxnet.jna.FunctionInfo.invoke(FunctionInfo.java:75)
at ai.djl.mxnet.engine.MxNDManager.invoke(MxNDManager.java:381)
at ai.djl.mxnet.engine.MxNDManager.invoke(MxNDManager.java:416)
at ai.djl.mxnet.engine.MxNDManager.randomNormal(MxNDManager.java:265)
at ai.djl.ndarray.NDManager.randomNormal(NDManager.java:1296)
at ai.djl.training.initializer.XavierInitializer.initialize(XavierInitializer.java:130)
at ai.djl.nn.Parameter.initialize(Parameter.java:207)
at ai.djl.nn.AbstractBaseBlock.initialize(AbstractBaseBlock.java:182)
at ai.djl.nn.SequentialBlock.initializeChildBlocks(SequentialBlock.java:224)
at ai.djl.nn.AbstractBaseBlock.initialize(AbstractBaseBlock.java:184)
at ai.djl.training.Trainer.initialize(Trainer.java:118)
at test.djl.mxnet.gpu.DjlMxnetGpu.main(DjlMxnetGpu.java:71)

--------- code --------------------------
`
public class DjlMxnetGpu {
public static void main(String[] args) throws IOException, TranslateException {
Engine engine = new MxEngineProvider().getEngine();
System.out.println("engine="+engine);
String engineName = engine.getEngineName();

	Device[] devices = engine.getDevices();
	for (Device d: devices)
	{
		System.out.println("engine device = "+d); 			
	}
	
	Device gpu = Device.gpu();
	System.out.println("my dev="+gpu);
			
	long inputSize = 28*28;
	long outputSize = 10;
	
	SequentialBlock block = new SequentialBlock();
	block.add(Blocks.batchFlattenBlock(inputSize));
	block.add(Linear.builder().setUnits(128).build());
	block.add(Activation::relu);
	block.add(Linear.builder().setUnits(64).build());
	block.add(Activation::relu);
	block.add(Linear.builder().setUnits(outputSize).build());
	
	Model model = Model.newInstance("mlp",gpu,engineName);
	System.out.println("model="+model);
	model.setBlock(block); 
	
	int batchSize = 40000;
	Mnist mnist = Mnist.builder().setSampling(batchSize, true).build();
	mnist.prepare(new ProgressBar());		
	
	DefaultTrainingConfig config = new DefaultTrainingConfig(Loss.softmaxCrossEntropyLoss())
		.addEvaluator(new Accuracy()) 
		.addTrainingListeners(TrainingListener.Defaults.logging());

	Trainer trainer = model.newTrainer(config);
	trainer.initialize(new Shape(1, 28 * 28));
	
	int epoch = 2;
	EasyTrain.fit(trainer, epoch, mnist, null);		
}

}`

---

std out msg :

engine=MXNet:1.8.0, capabilities: [
SIGNAL_HANDLER,
LAPACK,
BLAS_OPEN,
OPENMP,
OPENCV,
MKLDNN,
]
MXNet Library: C:\Users\java.djl.ai\mxnet\1.8.0-mkl-win-x86_64\mxnet.dll
engine device = cpu()
my dev=gpu(0)
model=Model (
Name: mlp
Data Type: float32
)
[11:59:02] C:\Users\Administrator\kimbergz\b4\src\imperative./imperative_utils.h:93: GPU support is disabled. Compile MXNet with USE_CUDA=1 to enable GPU support.
Exception in thread "main" ai.djl.engine.EngineException: MXNet engine call failed: MXNetError: GPU is not enabled
Stack trace:
File "..\src\resource.cc", line 167

at ai.djl.mxnet.jna.JnaUtils.checkCall(JnaUtils.java:1942)
at ai.djl.mxnet.jna.JnaUtils.imperativeInvoke(JnaUtils.java:521)
at ai.djl.mxnet.jna.FunctionInfo.invoke(FunctionInfo.java:75)
at ai.djl.mxnet.engine.MxNDManager.invoke(MxNDManager.java:381)
at ai.djl.mxnet.engine.MxNDManager.invoke(MxNDManager.java:416)
at ai.djl.mxnet.engine.MxNDManager.randomNormal(MxNDManager.java:265)
at ai.djl.ndarray.NDManager.randomNormal(NDManager.java:1296)
at ai.djl.training.initializer.XavierInitializer.initialize(XavierInitializer.java:130)
at ai.djl.nn.Parameter.initialize(Parameter.java:207)
at ai.djl.nn.AbstractBaseBlock.initialize(AbstractBaseBlock.java:182)
at ai.djl.nn.SequentialBlock.initializeChildBlocks(SequentialBlock.java:224)
at ai.djl.nn.AbstractBaseBlock.initialize(AbstractBaseBlock.java:184)
at ai.djl.training.Trainer.initialize(Trainer.java:118)
at test.djl.mxnet.gpu.DjlMxnetGpu.main(DjlMxnetGpu.java:71)

[iamuspace]

pom.xml

<dependencies>
    <dependency>
        <groupId>ai.djl</groupId>
        <artifactId>api</artifactId>
        <version>0.20.0</version>
    </dependency>
    <dependency>
        <groupId>ai.djl.mxnet</groupId>
        <artifactId>mxnet-native-auto</artifactId>
        <version>1.8.0</version>
    </dependency>
    <dependency>
        <groupId>ai.djl.mxnet</groupId>
        <artifactId>mxnet-engine</artifactId>
        <version>0.20.0</version>
    </dependency>
    <dependency>
        <groupId>ai.djl.mxnet</groupId>
        <artifactId>mxnet-native-cu102mkl</artifactId>
        <version>1.6.0</version>
        <classifier>win-x86_64</classifier>
    </dependency>
    <dependency>
        <groupId>ai.djl</groupId>
        <artifactId>basicdataset</artifactId>
        <version>0.20.0</version>
    </dependency>
</dependencies>

[frankfliu]

Can you provide your CUDA version?

A few thing to point out:

1. Please use BOM, it will avoid version mismatch between DJL packages
2. mxnet-engine-auto is deprecated, and should be been used any more
3. mxnet 1.6.0 is way too old, please use 1.9.1
4. Please use PyTorch if possible, PyTorch is much popular nowadays

Here is an exmaple of pom file:

<dependencyManagement>
    <dependencies>
        <dependency>
            <groupId>ai.djl</groupId>
            <artifactId>bom</artifactId>
            <version>0.21.0</version>
            <type>pom</type>
            <scope>import</scope>
        </dependency>
    </dependencies>
</dependencyManagement>
<dependencies>
    <dependency>
        <groupId>ai.djl</groupId>
        <artifactId>api</artifactId>
    </dependency>
    <dependency>
        <groupId>ai.djl.mxnet</groupId>
        <artifactId>mxnet-native-cu102mkl</artifactId>
        <classifier>win-x86_64</classifier>
    </dependency>
    <dependency>
        <groupId>ai.djl</groupId>
        <artifactId>basicdataset</artifactId>
    </dependency>
</dependencies>

[iamuspace]

Thank you so much again frankfliu.

The highest mxnet-native-cu with win-x86_64 is 1.6.0
so I used this.

ai.djl.mxnet
mxnet-native-cu102mkl
1.6.0
win-x86_64

I added BOM manager,
I have changed mxnet-native-cu without the label win-x86_64 but
same error occurs.

pom.xml is :



ai.djl
bom
0.21.0
pom
import





ai.djl
api
0.20.0


ai.djl.mxnet
mxnet-engine
0.20.0


ai.djl
basicdataset
0.20.0


ai.djl.mxnet
mxnet-native-cu112mkl
1.9.1

---

My Cuda version is 11.7

C:\Users\java>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Jun__8_16:59:34_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.7, V11.7.99
Build cuda_11.7.r11.7/compiler.31442593_0

---

I want to use JDL Cuda with any engine.
I will try again for Pytorch.
now I have error for pytorch engine.

[frankfliu]

MXNet doesn't have CU11.7 support, you have to install CUDA 11.2 to use MXNet.

mxnet-native-xxx, pytorch-native-xxx package is designed for offline distribution. If you just want to run it you don't need them. By default, DJL will detect your OS and download proper package for you. If DJL download CPU version of native package, that means there is no GPU native package for your system.