Fix special tests CI #7790
Description
🚀 CI / Tests
We currently have about 3 tests failing on master, it's a mix of NCCL error and "RuntimeError: Address already in use".
Need to find the reason for the failures and adjust CI to report it properly.
FAILED tests/plugins/test_deepspeed_plugin.py::test_deepspeed_multigpu_stage_2_accumulated_grad_batches[True]
FAILED tests/plugins/test_deepspeed_plugin.py::test_deepspeed_multigpu_stage_2_accumulated_grad_batches[False]
FAILED tests/plugins/test_sharded_plugin.py::test_ddp_sharded_plugin_manual_optimization[ddp_sharded_spawn]