pycauset.set_precision_mode

pycauset.set_precision_mode(mode: str) -> None

Set the current thread-local promotion precision mode.

Parameters

• mode: Either "lowest" or "highest".

Description

PyCauset has an explicit promotion policy that controls how result dtypes are chosen when mixing dtypes. This function selects the policy used by the current thread.

• "lowest": choose the smallest reasonable storage dtype (scale/storage-first default).
• "highest": choose the highest-precision storage dtype available within the participating operand ranks.

Notes:

• This controls storage dtype selection (the dtype of the resulting matrix/vector).
• It does not directly control accelerator internal compute dtype.

Example

import pycauset as pc

pc.set_precision_mode("highest")

See also

• pycauset.get_precision_mode
• pycauset.precision_mode
• pycauset.divide