Failed copying input tensor from /job:localhost/replica:0/task:0/device:CPU:0 to /job:localhost/replica:0/task:0/device:GPU:0 in order to run Identity: Dst tensor is not initialized. [Op:Identity]

Hi, I am trying DLS to understand how it works. I tried to train a model with RGB image. My GPU is RTX2080 with 8GB RAM.
Due to less GPU memory, my model can’t train in GPU. My model training works fine in cpu (I9-9900KF with 32GB RAM)

When i tried to use both CPU and GPU for training, after 1st epoch, it shows the following error.

"Failed copying input tensor from /job:localhost/replica:0/task:0/device:CPU:0 to /job:localhost/replica:0/task:0/device:GPU:0 in order to run Identity: Dst tensor is not initialized. [Op:Identity]"

Why i am getting this error? and how to solve it?

Note: as i said before, i am just trying this DLS tool. So i give small data set and basic model architecture. The batch size is set to 1 because of resource constrains.

I have attached the images of my model and settings

https://drive.google.com/drive/folders/1IyI2Js7bOc56g4IDjwusRWH0IUk93kyB?usp=sharing
(as a new user, i can embed only one object in the post. so i used Google drive. )

Hi @raguvaran,

Sorry, as of now DLS doesn’t support training on CPU &GPU at the same time. We would advise selecting either GPU or CPU at a time to train the model.

Hope this works for you

Thanks