NemoMegatron-aws-optimized docker missed libucc.so.1

[tangxianfeng]

The docker image cannot import torch because it removes hpcx in line 23, but seems never installs it back.

Error example:
python3 -c "import torch; print(torch.cuda.is_available())"
Traceback (most recent call last):
File "", line 1, in
File "/usr/local/lib/python3.10/dist-packages/torch/init.py", line 368, in
from torch._C import * # noqa: F403
ImportError: libucc.so.1: cannot open shared object file: No such file or directory

[tangxianfeng]

I found that if I comment out line 23 to 26, the docker will work. But I am not sure if there will be any other conflicts.

[mhuguesaws]

Thank you for reporting. Our team is focus adding auto testing on FSDP at this time and will move to megatron next.
Nemo is third in our queue.