Training on multiple nodes

[Singh__Jagraj]

How to train a NN on multiple nodes using different parameters.

[ChithraJ_Intel]

Hi Jagraj, Thanks for reaching out to us. Could you please elaborate your issue.Are you trying to submit multiple single-node jobs by giving different arguments using -F flag or are you looking for distributed multi-node training of a neural network using distributed training frameworks such as Horovod.

[Singh__Jagraj]

i want to submit multiple single node jobs with different arguments. 

[ChithraJ_Intel]

Hi Jagraj,

We tried to run a sample keras code which takes learning rate as argument & submitted it as multiple jobs with different values for the learning rate.We could see that it's working fine without any errors. Could you please follow the below steps which we performed:
Step 1:

source activate tensorflow

[Note : tensorflow is the preinstalled conda environment in Devcloud which comes along with oneAPI]
Step 2: Install the required packages in the environmet:

 pip install keras --user

Step3 : Submit the job script by passing different values for learning rate as follows:

qsub myjob -F "0.1"
qsub myjob -F "0.01"

[Note: Job file and python script are attached]

Hopes this helps for you.

[ChithraJ_Intel]

Hi Jagraj,

Could you please let us know the solution provided answer your query. 

[Singh__Jagraj]

Thanks for the solution.

[ChithraJ_Intel]

Hi Jagraj,

Can we close this case if your issue got resolved?

[ChithraJ_Intel]

Hi Jagraj,

We are closing this case since we were able to provide solution.Please feel free to raise a new thread if you have further issues.