Types¶

mujou.types ¶

The types of mujou that annotations use.

Size is a shape: an int n is (n,), and () has no dimension, as the batch of mujou.pc (mujou.pc.msg.activity_spec). Unlike torch.Size, it takes an int as well.
Device is where a node computes or a tensor is made: a torch.device or its name, such as "cpu" or "cuda:0".
Message is a message the host writes: a torch.Tensor, a Python scalar (int, float, bool or Timestamp), or a dict of them (a TypedDict of the port's DictSpec).

Import them from here::

from mujou.types import Device, Message, Size

Size = int | tuple[int, ...]

Device = device | str

Message = Tensor | int | float | bool | Timestamp | Mapping[str, object]