[BUG] OSError: /usr/local/lib/python3.10/dist-packages/fbgemm_gpu/fbgemm_gpu_config.so: undefined symbol against PyTorch 2.7

Add Link

https://pytorch.org/tutorials/intermediate/pinmem_nonblock.html

Describe the bug

Tutorial failing with the following error:

Unexpected failing examples:
/var/lib/workspace/intermediate_source/pinmem_nonblock.py failed leaving traceback:
Traceback (most recent call last):
  File "/var/lib/workspace/intermediate_source/pinmem_nonblock.py", line 642, in <module>
    from tensordict import TensorDict
  File "/usr/local/lib/python3.10/dist-packages/tensordict/__init__.py", line 6, in <module>
    import tensordict._reductions
  File "/usr/local/lib/python3.10/dist-packages/tensordict/_reductions.py", line 11, in <module>
    from tensordict._lazy import LazyStackedTensorDict
  File "/usr/local/lib/python3.10/dist-packages/tensordict/_lazy.py", line 37, in <module>
    from tensordict.memmap import MemoryMappedTensor
  File "/usr/local/lib/python3.10/dist-packages/tensordict/memmap.py", line 22, in <module>
    from tensordict.utils import _shape, implement_for, IndexType, NESTED_TENSOR_ERR
  File "/usr/local/lib/python3.10/dist-packages/tensordict/utils.py", line 94, in <module>
    from torchrec import KeyedJaggedTensor
  File "/usr/local/lib/python3.10/dist-packages/torchrec/__init__.py", line 10, in <module>
    import torchrec.distributed  # noqa
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/__init__.py", line 38, in <module>
    from torchrec.distributed.model_parallel import DistributedModelParallel  # noqa
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/model_parallel.py", line 26, in <module>
    from torchrec.distributed.planner import EmbeddingShardingPlanner, Topology
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/planner/__init__.py", line 24, in <module>
    from torchrec.distributed.planner.planners import EmbeddingShardingPlanner  # noqa
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/planner/planners.py", line 21, in <module>
    from torchrec.distributed.planner.constants import BATCH_SIZE, MAX_SIZE
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/planner/constants.py", line 12, in <module>
    from torchrec.distributed.embedding_types import EmbeddingComputeKernel
  File "/usr/local/lib/python3.10/dist-packages/torchrec/distributed/embedding_types.py", line 16, in <module>
    from fbgemm_gpu.split_table_batched_embeddings_ops_training import EmbeddingLocation
  File "/usr/local/lib/python3.10/dist-packages/fbgemm_gpu/__init__.py", line 71, in <module>
    _load_library(f"{library}.so")
  File "/usr/local/lib/python3.10/dist-packages/fbgemm_gpu/__init__.py", line 21, in _load_library
    raise error
  File "/usr/local/lib/python3.10/dist-packages/fbgemm_gpu/__init__.py", line 17, in _load_library
    torch.ops.load_library(os.path.join(os.path.dirname(__file__), filename))
  File "/var/lib/ci-user/.local/lib/python3.10/site-packages/torch/_ops.py", line 1392, in load_library
    ctypes.CDLL(path)
  File "/usr/lib/python3.10/ctypes/__init__.py", line 374, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: /usr/local/lib/python3.10/dist-packages/fbgemm_gpu/fbgemm_gpu_config.so: undefined symbol: _ZN5torch3jit17parseSchemaOrNameERKSsb

build log

Please submit fixes against the 2.7-RC-TEST branch and enable in .jenkins/validate_tutorials_built.py.

Describe your environment

CUDA: 12.6
PyTorch: 2.7

cc @vmoens @nairbv

Start with intermediate_source/pinmem_nonblock.py and reproduce its failure using the linked build log, focusing on the PyTorch 2.7 and fbgemm_gpu import error. Then inspect .jenkins/validate_tutorials_built.py and the 2.7-RC-TEST branch requirements. Done means the tutorial validation completes successfully with PyTorch 2.7 and the validation is enabled as requested.