Hello, I am getting an error running the chapter two notebook. I am using paperspace’s gradient service to run all of this with a P5000 GPU. Just under the Training your model, and using it to clean your data subtitle, the code block
learn = cnn_learner(dls, resnet18, metrics=error_rate)
learn.fine_tune(4)
looks like it is running ok at first and prints out the table with the epoch and such. After about 10-30 seconds I get the error
RuntimeError: DataLoader worker (pid 14789) is killed by signal: Killed.
I have seen a similar error on forums but not with this notebook. The solution there was to set num_workers = 0 in the dls. Doing this causes the kernal to crash every time. Has anyone else had this issue?
I had the same problem, but turns out I didn’t run my notebook on a GPU.
In paperspace’s gradient, before you start the notebook, make sure you choose the GPU, sometimes, they default to free CPU.
Oh, you know what you are right. I had made the notebook instance with the free gpu initially but now it is giving me the free cpu option. Currently, it says there are no free gpu instances available, but for future reference is there a way to change what my notebook is using without creating a whole new one?
You can change the setting of the notbook at anytime under the menu “Runtime” --> “Change runtime type”. There you can select None or GPU. And also under “Runtime” you can cancel or request a new session.