Error using parallel-cpu with trainnet function

Hi, I got this error when trying to train a neural network using the trainnet function. The first time I trained the network everything works fine, but when I tried to use validation data I got an error. Then I tried to retrain the network without validation data but I got this error:

Error using deep.internal.train.Trainer/train (line 74)

Error detected on worker 3.

Error in deep.internal.train.trainnet (line 54)

net = train(trainer, net, mbq);

Error in trainnet (line 42)

[net,info] = deep.internal.train.trainnet(mbq, net, loss, options, ...

Caused by:

Error using deep.internal.train.ParallelTrainer/computeTraining (line 101)

Out of Memory during deserialization

I have deleted the Jobs of the cluster but nothing worked.

Can anybody help me to solve the issue?

[Saurabh]

Hi @Ramiro,

Could you please share the details regarding the neural network and the memory allocation per worker? The minimum required memory is 4GB, with 8GB recommended. To optimize performance, consider increasing the memory available to each worker, which can typically be achieved by running fewer workers per compute node.

[Ramiro]

Hi @Saurabh, I have attached the files of the source code.

Regards

Ramiro

[Subhajyoti]

Hi @Ramiro,

Also share the train and test MAT files.

They are required to run the script.

Answer 1

[Sivsankar]

Hi @Ramiro,

Without the ability to execute the scripts, it is challenging to pinpoint the exact solution to this error. However, I can offer some troubleshooting steps that may assist you.

As indicated by the error message, this could be a memory-related issue concerning the workers. Ensure that sufficient memory is allocated per core on the cluster. Since the error occurred during network retraining, it is important to verify that the allocated memory is properly cleared.

Additionally, the error may suggest that the GPU lacks adequate memory to complete the network training. To mitigate memory requirements, consider decreasing the "MiniBatchSize" training option or reducing the size of your training dataset. Also, ensure that the GPU memory is cleared before initiating retraining.

I hope these suggestions prove helpful.