Subprocess failed (exit code: 139)

I don't know whether it's a common problem.
When I imported some big deep-learning module, for example, paddlenlp, and after I finished some computation, nRepl would crash with the info 'Subprocess failed (exit code: 139)'.
However, when the computation was running, it seemed that nRepl wouldn't crash. When I repeated a computation for 1000 times, and each time I modified a bit to make it acctually run so many iterations, nRepl would crash after the loop finished.
But even when I didn't do anything after I require-python this module, it would also crash. Sometimes, nRepl didn't crash after a series of computations, and even an hour passed it wouldn't crash any more if no operation inputted, until I ran another command (even clojure-native form) on it.
I used Java 1.8 and python 3.7, cider on emacs, and it repeated when I changed into python 3.10.

No source files, tests, or entry points are named. Start by reproducing the crash with libpython-clj, nREPL, Java 1.8, and Python 3.7 or 3.10 while importing paddlenlp and running the reported computations; done means the exit-code-139 failure is explained and no longer occurs in those scenarios.