Playing a bit more with profila + pytensor, I have a graph that works fine when the jupyter extension is not loaded, but fails to evaluate when it is loaded, with this cryptic error at the end of the stack trace:
RuntimeError: LLVM IR parsing error
<string>:2961:36: error: expected unsigned integer
!47 = !DILocation(column: 1, line: -1, scope: !30)
It's preceded by this warning
[/tmp/tmpc42ipk57:3](http://localhost:8888/tmp/tmpc42ipk57#line=2): NumbaDebugInfoWarning: Could not find source for function: <function __numba_array_expr_0x7f9889c61eb0 at 0x7f988d56ef20>. Debug line information may be inaccurate.
tensor_variable = cholesky(cov)
Cholesky is one of this strange Ops that are implemented as a pointer to the lapack function
Playing a bit more with profila + pytensor, I have a graph that works fine when the jupyter extension is not loaded, but fails to evaluate when it is loaded, with this cryptic error at the end of the stack trace:
It's preceded by this warning
Cholesky is one of this strange Ops that are implemented as a pointer to the lapack function