sima_utils.transformer.utils
============================

.. py:module:: sima_utils.transformer.utils


Attributes
----------

.. autoapisummary::

   sima_utils.transformer.utils.mla_row_size
   sima_utils.transformer.utils.mla_num_tiles
   sima_utils.transformer.utils.mla_max_num_rows_per_tile
   sima_utils.transformer.utils.mla_max_num_rows


Functions
---------

.. autoapisummary::

   sima_utils.transformer.utils.ceil_div
   sima_utils.transformer.utils.ceil_div_row
   sima_utils.transformer.utils.round_up_to
   sima_utils.transformer.utils.round_up_to_row
   sima_utils.transformer.utils.calc_freq_real_imag


Module Contents
---------------

.. py:data:: mla_row_size
   :type:  int
   :value: 16


.. py:data:: mla_num_tiles
   :type:  int
   :value: 100


.. py:data:: mla_max_num_rows_per_tile
   :type:  int
   :value: 4096


.. py:data:: mla_max_num_rows
   :type:  int
   :value: 409600


.. py:function:: ceil_div(x: int, y: int) -> int

.. py:function:: ceil_div_row(x: int) -> int

.. py:function:: round_up_to(x: int, y: int) -> int

.. py:function:: round_up_to_row(x: int) -> int

.. py:function:: calc_freq_real_imag(max_num_tokens: int, rope_type: str, theta: float, head_dim: int, scaling_cfg: dict, idx_base: int = 0) -> tuple[numpy.ndarray, numpy.ndarray]