Recently, I encountered a very strange problem in the process of deploying AI services.
When deploying the AI model to the centerOS server, the system reports an "Illegal command (core dump)" error. The model can run through the pytorch 1.6 version on windows and ubuntu systems, and the code itself is no problem. I have no choice but to check one by one, and finally found that when loading a method in torch, it was directly interrupted, and the underlying libtorch_cpu.so reported an error. It is suspected that the system and torch version are not compatible.
Reduce the torch version to version 1.5, and finally solve it. . . However, some APIs in version 1.5 are not available, and you need to write them yourself, which is a bit tricky. It seems that it is better to use the ubuntu system for the server deployed by ai, which is more stable and avoids various strange errors. Mark it.