opaque

Bases: Tensor

data

Very important to keep access to all special attr of OpaqueTensor.

to_base_tensor()

Wrap _to_base_tensor with jit export infos.

OpaqueTensorRef(meta_tensor: torch.Tensor, opaque_tensor: OpaqueTensor)

Bases: Tensor

Allow to pass through 'tracing'.

find_opaque_ref_by_py_id(module: torch.nn.Module, py_id: int)

Allow to fetch back the opaque parameter once passed the jit 'wall'.

opaque_to_final_tensor(rtensor: torch.Tensor) -> torch.Tensor

Even if OpaqueTensor are composed it exposes fully expanded tensor.

So for example: an OffloadedTensor that contains a QTensor will 'load' then 'decompress' to show final fp tensor.

set_opaque_tensor_in_params_as_ref(model: torch.nn.Module)

Transform OpaqueTensor Parameters into OpaqueTensorRef.

This is applied at export time of torch_to_nnef Just before doing any tracing