nvalchemi.distributed.AdapterStatus#

class nvalchemi.distributed.AdapterStatus(kind, target, state, install_site='', error=None)[source]#

Introspectable record of one adapter’s lifecycle state.

Returned by AdapterRegistry.list_active(). Surfaces what’s been swapped in this process — useful when debugging “why is fairchem behaving weirdly outside the distributed scope” kinds of questions.

Parameters:
  • kind (Literal['op', 'jit', 'python', 'method'])

  • target (str)

  • state (Literal['pending', 'installed', 'restored', 'failed'])

  • install_site (str)

  • error (str | None)

kind#

"op" for OpAdapter, "jit" / "python".

Type:

Literal[‘op’, ‘jit’, ‘python’, ‘method’]

target#

Human-readable identifier of the adapted callable ("torch.ops.fairchem._kernel_xyz" / "aimnet.nbops.mol_sum").

Type:

str

state#

"pending" (registered but not installed), "installed", "restored" (installed then cleaned up), "failed" (install raised — see error).

Type:

Literal[‘pending’, ‘installed’, ‘restored’, ‘failed’]

install_site#

"filename:lineno" capturing where the adapter was constructed. Empty when not auto-captured.

Type:

str

error#

str(exception) when state == "failed"; else None.

Type:

str | None