angle_emb.utils
===============

.. py:module:: angle_emb.utils


Attributes
----------

.. autoapisummary::

   angle_emb.utils.logger


Functions
---------

.. autoapisummary::

   angle_emb.utils.cosine_similarity
   angle_emb.utils.set_device
   angle_emb.utils.find_all_linear_names
   angle_emb.utils.get_pooling


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

.. py:data:: logger

.. py:function:: cosine_similarity(vec1: List[int], vec2: List[int])

   Calculate cosine similarity between two vectors.

   :param vec1: a list of integers
   :param vec2: a list of integers
   :return: a float value between 0 and 1, indicating the similarity between the two vectors.


.. py:function:: set_device() -> str

   Set device automatically

   :return: str, device name


.. py:function:: find_all_linear_names(model: transformers.PreTrainedModel, linear_type: Optional[object] = None) -> List[str]

   Find all linear layer names

   :param model: PreTrainedModel
   :param linear_type: Optional[object] = None, linear type, such as nn.Linear and bnb.nn.Linear4bit.

   :return: List[str], linear layer names


.. py:function:: get_pooling(outputs: torch.Tensor, inputs: Dict, pooling_strategy: str, padding_side: str) -> torch.Tensor

   Pooling the model outputs.

   :param outputs:  torch.Tensor. Model outputs (without pooling)
   :param inputs:  Dict. Model inputs
   :param pooling_strategy:  str.
       Pooling strategy [`cls`, `cls_avg`, `cls_max`, `last`, `avg`, `mean`, `max`, `all`, int]
   :param padding_side:  str. Padding strategy of tokenizers (`left` or `right`).
       It can be obtained by `tokenizer.padding_side`.