Torch 2.7.0 nightly cuda 12.6 and cuda 12.8 builds are broken on Amazon linux 2023 #148120
Description
🐛 Describe the bug
Failure can be seen here:
https://github.com/pytorch/test-infra/actions/runs/13558218752/job/37934127476
Error:
2025-02-27T16:00:09.7113764Z + python3 .ci/pytorch/smoke_test/smoke_test.py --package torchonly
2025-02-27T16:00:09.7114301Z Traceback (most recent call last):
2025-02-27T16:00:09.7114936Z File "/pytorch/pytorch/.ci/pytorch/smoke_test/smoke_test.py", line 11, in <module>
2025-02-27T16:00:09.7115416Z import torch
2025-02-27T16:00:09.7115840Z File "/usr/local/lib64/python3.9/site-packages/torch/__init__.py", line 401, in <module>
2025-02-27T16:00:09.7116346Z from torch._C import * # noqa: F403
2025-02-27T16:00:09.7116835Z ImportError: libcufile.so.0: cannot open shared object file: No such file or directory
2025-02-27T16:00:09.7117696Z File "/home/ec2-user/actions-runner/_work/test-infra/test-infra/test-infra/.github/scripts/run_with_env_secrets.py", line 102, in <module>
2025-02-27T16:00:09.7118364Z main()
2025-02-27T16:00:09.7118954Z File "/home/ec2-user/actions-runner/_work/test-infra/test-infra/test-infra/.github/scripts/run_with_env_secrets.py", line 98, in main
2025-02-27T16:00:09.7119687Z run_cmd_or_die(f"docker exec -t {container_name} /exec")
2025-02-27T16:00:09.7120450Z File "/home/ec2-user/actions-runner/_work/test-infra/test-infra/test-infra/.github/scripts/run_with_env_secrets.py", line 39, in run_cmd_or_die
2025-02-27T16:00:09.7121360Z raise RuntimeError(f"Command {cmd} failed with exit code {exit_code}")
Looks like this is result of cufile addition: #145748
Versions
torch-2.7.0.dev20250227+cu126