emmi.modules.decoder.deep_perceiver_decoder
===========================================

.. py:module:: emmi.modules.decoder.deep_perceiver_decoder

.. autoapi-nested-parse::

   Contains a PerceiverDecoder implementation.



Classes
-------

.. autoapisummary::

   emmi.modules.decoder.deep_perceiver_decoder.DeepPerceiverDecoder


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

.. py:class:: DeepPerceiverDecoder(config)

   Bases: :py:obj:`torch.nn.Module`


   A deep Perceiver decoder module. Can be configured with different number of layers and hidden dimensions.
   However, it should be noted that this layer is not a full-fledged Perceiver, since it only has a cross-attention mechanism.

   Initialize the DeepPerceiverDecoder.

   :param config: Configuration for the DeepPerceiverDecoder module.


   .. py:attribute:: blocks


   .. py:method:: forward(kv, queries, unbatch_mask = None, attn_kwargs = None, condition = None)

      Forward pass of the model.

      :param x: Latent tokens as dense tensor (batch_size, num_latent_tokens, dim).
      :param pos: Query positions (batch_size, num_output_pos, pos_dim).
      :param block_kwargs: Additional arguments for the block.
      :param unbatch_mask: Unbatch mask.

      :returns: The predictions as sparse tensor (batch_size * num_output_pos, num_out_values).