Model creation breaks on dynamic input #4108
Description
🐛 Bug
The following PR is being tested to improve dynamic shape functionality on simple models on TPU v3.
I see the error below at y_pred = model(x_test) when x_test = torch.nonzero(x_test) is included in the code sequence - without which the code runs normally.
https://symbolize.stripped_domain/r/?trace=7f400b6b400b,7f400b6b408f,7f400b49d23f&map=
*** SIGABRT received by PID 393548 (TID 393548) on cpu 25 from PID 393548; stack trace: ***
PC: @ 0x7f400b6b400b (unknown) raise
@ 0x7f3ea567f2d4 1120 (unknown)
@ 0x7f400b6b4090 919423072 (unknown)
@ 0x7f400b49d240 (unknown) (unknown)
https://symbolize.stripped_domain/r/?trace=7f400b6b400b,7f3ea567f2d3,7f400b6b408f,7f400b49d23f&map=13dcea8db75a2ed88ce9356603629dcb:7f3e95b07000-7f3ea591c9e0
E1020 01:32:56.563561 393548 coredump_hook.cc:395] RAW: Remote crash data gathering hook invoked.
E1020 01:32:56.563569 393548 coredump_hook.cc:441] RAW: Skipping coredump since rlimit was 0 at process start.
E1020 01:32:56.563576 393548 client.cc:243] RAW: Coroner client retries enabled (b/136286901), will retry for up to 30 sec.
E1020 01:32:56.563580 393548 coredump_hook.cc:502] RAW: Sending fingerprint to remote end.
E1020 01:32:56.563585 393548 coredump_socket.cc:120] RAW: Stat failed errno=2 on socket /var/google/services/logmanagerd/remote_coredump.socket
E1020 01:32:56.563592 393548 coredump_hook.cc:506] RAW: Cannot send fingerprint to Coroner: [NOT_FOUND] Missing crash reporting socket. Is the listener running?
E1020 01:32:56.563595 393548 coredump_hook.cc:580] RAW: Discarding core.
E1020 01:32:56.819510 393548 process_state.cc:775] RAW: Raising signal 6 with default behavior
Aborted (core dumped)