Closed james20141606 closed 3 years ago
seems that the bug is caused by a wrong torch version. If I don't have pytorch installed first, pip install sparseml
will install pytorch with version 1.8.1+cu102
which is not compatible with my cuda version and sparseml's requirement. If I install appropriate pytorch first the bug won't appear
edit: it seems that the above could not solve the problem entirely. I could still got the same error.
Hi @james20141606, thank you for bringing this up! There definitely is an issue where PyTorch should not be installing with 1.8. We'll be sure to get that fixed ASAP, that is unrelated to this, though (thank you for finding that!).
This issue will be with the DeepSparse engine and a hardware dependency. I'm going to move this over to the DeepSparse repository and we'll have someone on the runtime team reach out for more info on the environment it's being run on.
thanks for the help. seems that it is unrelated to pytorch version. I agree it should be a hardware dependency issue. seems that deepspare engine might not be compatible with some CPUs/GPUs.
Hi @james20141606 , can you supply some more details about the system you are running on? e.g. OS version/distribution, CPU version and contents of /proc/cpuinfo (if possible). It would also help to know if you are running inside a VM and if so, which VM software is being used.
Hello @james20141606 Let us know if the issue is still happening and you are able to share some more information per the last message? If we don't hear back, we'll close out this issue in the next few weeks as we'll assume you're all set. Thank you.
Cheers, Jeannie / Neural Magic
Hi @james20141606 As there has been no response, we'll close this issue but please re-open it if you're able to provide some more details per our last few comments. Thank you. Regards, Jeannie / Neural Magic
Hi, I tried to run the following codes and it seems that it could run smoothly on my mac/terminal, but always died if I run in jupyter notebook:
The error information:
version: I tried to install sparseml using
pip install sparseml
, and it will install torch with version1.8.1+cu102
(which I found strange since the doc said sparseml requires <=1.8.0). I also tried to downgrade torch to 1.8.0 but the same error still happens. The error appears both on CPU or GPU.