
    iB                        d dl Z d dlmZ d dlmZ d dlZd dlmZ ddlmZ	 ddl
mZ ddlmZmZ dd	lmZ dd
lmZmZ ddlmZmZ ddlmZ ddlmZmZmZmZ ddlmZm Z  ddl!m"Z"m#Z# e ed       G d de                    Z$ G d dejJ                        Z& G d dejJ                        Z' G d dejJ                        Z(d Z) ed      d<d       Z*dejV                  de,d ejV                  fd!Z-	 d=d"ejJ                  d#ejV                  d$ejV                  d%ejV                  d&ejV                  dz  d'e.d(e.d)ee   fd*Z/ ee*       G d+ d,ejJ                               Z0 G d- d.ejJ                        Z1 G d/ d0e      Z2e G d1 d2e             Z3 ed3       G d4 d5e3             Z4e G d6 d7e             Z5 ed8       G d9 d:e3             Z6g d;Z7y)>    N)Callable)	dataclass)nn   )initialization)ACT2FN)use_kernel_func_from_hubuse_kernelized_func)GradientCheckpointingLayer)BaseModelOutputCausalLMOutput)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)ModelOutputTransformersKwargsauto_docstringcan_return_tuple)check_model_inputsmaybe_autocast   )ParakeetCTCConfigParakeetEncoderConfigz
    Extends [~modeling_outputs.BaseModelOutput] to include the output attention mask since sequence length is not preserved in the model's forward.
    )custom_introc                   6    e Zd ZU dZej
                  dz  ed<   y)ParakeetEncoderModelOutputNattention_mask)__name__
__module____qualname__r   torchTensor__annotations__     x/home/obispo/Crisostomo_bridge/mision_env/lib/python3.12/site-packages/transformers/models/parakeet/modeling_parakeet.pyr   r   (   s     +/NELL4'.r%   r   c                        e Zd ZU dZej
                  ed<   ddef fdZ ej                         dej
                  fd       Z
 xZS )$ParakeetEncoderRelPositionalEncodingz*Relative positional encoding for Parakeet.inv_freqconfigc                 6   t         |           |j                  | _        d}d|t        j                  d|j
                  dt        j                        j                  |t        j                        |j
                  z  z  z  }| j                  d|d	       y )
N     @      ?r      dtype)devicer0   r)   F)
persistent)
super__init__max_position_embeddingsr!   arangehidden_sizeint64tofloatregister_buffer)selfr*   r1   baser)   	__class__s        r&   r4   z-ParakeetEncoderRelPositionalEncoding.__init__7   s    '-'E'E$Q 2 2AU[[ILLTZbgbmbmLn$$%
 	ZeDr%   hidden_statesc                    |j                   d   }|| j                  kD  rt        d| d| j                   d      t        j                  |dz
  | d|j
                        }| j                  d d d d f   j                         j                  |j                   d   dd      j                  |j
                        }|d d d d f   j                         }t        |j
                  j                  t              r/|j
                  j                  dk7  r|j
                  j                  nd	}t        |d
      5  |j                         |j                         z  j                  dd      }|j                         }|j!                         }	t        j"                  ||	gd      }
 |
j$                  g |
j                   d d d }
d d d        
j                  |j&                        S # 1 sw Y   %xY w)Nr   zSequence Length: z= has to be less or equal than config.max_position_embeddings .r1   r   mpscpuF)device_typeenabledr.   dimr/   )shaper5   
ValueErrorr!   r6   r1   r)   r:   expandr9   
isinstancetypestrr   	transposesincosstackreshaper0   )r<   r?   
seq_lengthposition_idsinv_freq_expandedposition_ids_expandedrF   freqsrR   rS   	pos_embeds              r&   forwardz,ParakeetEncoderRelPositionalEncoding.forwardE   s   "((+
444#J< 02262N2N1OqR 
 ||JNZKML`L`aMM$4-(..0778K8KA8NPRTUVYYZgZnZno 	 !-T4] ; A A C -..33S9m>R>R>W>W[`>`   %% 	
 UC 	E&,,.1F1L1L1NNYYZ[]^_E))+C))+CS#JB7I)	))D9??3B+?DDI	E ||-"5"5|66	E 	Es   5BG%%G.N)r   r   r    __doc__r!   r"   r#   r   r4   no_gradr\   __classcell__r>   s   @r&   r(   r(   2   sF    4llE4 E U]]_7U\\ 7 7r%   r(   c                   *     e Zd Zdef fdZd Z xZS )ParakeetEncoderFeedForwardr*   c                 `   t         |           t        j                  |j                  |j
                  |j                        | _        t        |j                     | _
        t        j                  |j
                  |j                  |j                        | _        |j                  | _        y )Nbias)r3   r4   r   Linearr7   intermediate_sizeattention_biaslinear1r   
hidden_act
activationlinear2activation_dropoutr<   r*   r>   s     r&   r4   z#ParakeetEncoderFeedForward.__init__e   s|    yy!3!3V5M5MTZTiTij !2!23yy!9!96;M;MTZTiTij"(";";r%   c                     | j                  | j                  |            }t        j                  j	                  || j
                  | j                        }| j                  |      }|S )Nptraining)rl   rj   r   
functionaldropoutrn   rs   rm   )r<   r?   s     r&   r\   z"ParakeetEncoderFeedForward.forwardl   sU    ](CD--mt?V?Vaeanan-o]3r%   r   r   r    r   r4   r\   r`   ra   s   @r&   rc   rc   d   s    <4 <r%   rc   c                   .     e Zd Zddef fdZddZ xZS ) ParakeetEncoderConvolutionModuler*   c           	      6   t         |           |j                  }|&|j                  }t        t        |dd         | _        n#|d   }t        |j                  dd         | _        |dz
  dz  | _        t        j                  |d|z  ddd|j                  	      | _        t        j                  |||d| j                  ||j                  
      | _        t        j                  |      | _        t        j                  ||ddd|j                  	      | _        y)z
        Args:
            config (ParakeetEncoderConfig): Configuration for the model.
            module_config (dict): Configuration for the module (e.g., encoder or decoder).
        Nrk   silukernel_sizerl   r   r.   r   )r{   stridepaddingrf   )r|   r}   groupsrf   )r3   r4   r7   conv_kernel_sizer   getattrrl   getr}   r   Conv1dconvolution_biaspointwise_conv1depthwise_convBatchNorm1dnormpointwise_conv2)r<   r*   module_configchannelsr{   r>   s        r&   r4   z)ParakeetEncoderConvolutionModule.__init__t   s    	%%  11K$WV\6%JKDO'6K$]%6%6|V%LMDO#aA-!yya(l!QVMdMd 
 !iiLL((
 NN8,	!yyhAaI`I` 
r%   c                     |j                  dd      }| j                  |      }t        j                  j	                  |d      }|c|j
                  t        j                  k(  rt        j                  | d      }nt        j                  |dk(   d      }|j                  |d      }| j                  |      }| j                  |      }| j                  |      }| j                  |      }|j                  dd      S )aY  
        Compute convolution module.

        Args:
            hidden_states (`torch.Tensor` of shape `(batch, time, channels)`): Input tensor.
            attention_mask (`torch.Tensor` of shape `(batch, 1, time, time)`): Attention mask.

        Returns:
            `torch.Tensor`: Output tensor of shape `(batch, time, channels)`.

        r   r.   rH           )rQ   r   r   rt   glur0   r!   boolallmasked_fillr   r   rl   r   )r<   r?   r   all_masked_rowss       r&   r\   z(ParakeetEncoderConvolutionModule.forward   s     &//15 ,,];))-Q)? %##uzz1"'))^O"C"'))n.C,D!"L)55osKM ++M:		-06,,];&&q!,,r%   r]   rv   ra   s   @r&   rx   rx   s   s     
4  
D"-r%   rx   c                     | dd| j                   d   dz  f   }| d| j                   d   dz  df   }t        j                  | |fd      S )z*Rotates half the hidden dims of the input..NrB   r.   rH   )rK   r!   cat)xx1x2s      r&   rotate_halfr      sZ    	
3"!''"+"""	#B	
3q ""	#B99rc2YB''r%   rotary_pos_embc                     |j                  |      }|j                  |      }| |z  t        |       |z  z   }||z  t        |      |z  z   }||fS )a  Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q (`torch.Tensor`): The query tensor.
        k (`torch.Tensor`): The key tensor.
        cos (`torch.Tensor`): The cosine part of the rotary embedding.
        sin (`torch.Tensor`): The sine part of the rotary embedding.
        unsqueeze_dim (`int`, *optional*, defaults to 1):
            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
    Returns:
        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
    )	unsqueezer   )qkrS   rR   unsqueeze_dimq_embedk_embeds          r&   apply_rotary_pos_embr      sY    & --
&C
--
&C3w;q>C/0G3w;q>C/0GGr%   r?   n_repreturnc                     | j                   \  }}}}|dk(  r| S | dddddddddf   j                  |||||      } | j                  |||z  ||      S )z
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    r   N)rK   rM   rU   )r?   r   batchnum_key_value_headsslenhead_dims         r&   	repeat_kvr      so    
 2?1D1D.Ehz!!Qa"23::5BUW\^bdlmM  (;e(CT8TTr%   modulequerykeyvaluer   scalingru   kwargsc                 T   t        || j                        }t        || j                        }	t        j                  ||j	                  dd            |z  }
|#|d d d d d d d |j
                  d   f   }|
|z   }
t        j                  j                  |
dt        j                        j                  |j                        }
t        j                  j                  |
|| j                        }
t        j                  |
|	      }|j	                  dd      j                         }||
fS )Nr.   r   rJ   rB   rI   r0   rq   r   )r   num_key_value_groupsr!   matmulrQ   rK   r   rt   softmaxfloat32r9   r0   ru   rs   
contiguous)r   r   r   r   r   r   ru   r   
key_statesvalue_statesattn_weightscausal_maskattn_outputs                r&   eager_attention_forwardr      s    3 ; ;<JUF$?$?@L<<z';';Aq'ABWLL!$Q1.D
0@0@0D.D%DE#k1==((2U]](SVVW\WbWbcL==((6??([L,,|\:K''1-88:K$$r%   c                        e Zd ZdZdedef fdZ	 ddej                  dej                  dz  dej                  dz  d	e	e
   d
eej                  ej                  f   f
dZd Z xZS )ParakeetEncoderAttentionztMulti-head attention with relative positional encoding. See section 3.3 of https://huggingface.co/papers/1901.02860.r*   	layer_idxc                    t         |           || _        || _        t	        |d|j
                  |j                  z        | _        |j                  |j                  z  | _	        | j                  dz  | _
        |j                  | _        d| _        t        j                  |j
                  |j                  | j                  z  |j                        | _        t        j                  |j
                  |j                  | j                  z  |j                        | _        t        j                  |j
                  |j                  | j                  z  |j                        | _        t        j                  |j                  | j                  z  |j
                  |j                        | _        t        j                  |j
                  |j                  | j                  z  d      | _        t        j*                  t-        j.                  |j                  | j                              | _        t        j*                  t-        j.                  |j                  | j                              | _        y )Nr   g      Fre   )r3   r4   r*   r   r   r7   num_attention_headsr   r   r   r   attention_dropout	is_causalr   rg   ri   q_projk_projv_projo_projrelative_k_proj	Parameterr!   zerosbias_ubias_vr<   r*   r   r>   s      r&   r4   z!ParakeetEncoderAttention.__init__  s   "
F4F4F&JdJd4de$*$>$>&B\B\$\!}}d*!'!9!9ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii&&68J8JQWQfQf
  "yy););V=W=WZ^ZgZg=gnstll5;;v/I/I4==#YZll5;;v/I/I4==#YZr%   Nr?   position_embeddingsr   r   r   c           
         |j                   d d }|\  }}||d| j                  f}| j                  |      j                  |      j	                  dd      }	| j                  |      j                  |      j	                  dd      }
| j                  |      j                  |      j	                  dd      }t        j                  | j                  j                  t              }|	| j                  j                  d| j                  j                  d| j                        z   }|	| j                  j                  d| j                  j                  d| j                        z   }| j                  |      }|j                  |d| j                  j                  | j                        }||j!                  dddd      z  }| j#                  |      }|dd |f   }|| j$                  z  }|)|j'                  |j)                         t+        d            } || f||
||| j,                  sdn| j.                  | j$                  d	|\  }} |j0                  g |d j3                         }| j5                  |      }||fS )
NrB   r   r.   r   r   .z-infr   )r   r   r   r   ru   r   )rK   r   r   viewrQ   r   r   r   get_interfacer*   _attn_implementationr   r   r   r   r   permute
_rel_shiftr   masked_fill_logical_notr:   rs   r   rU   r   r   )r<   r?   r   r   r   input_shape
batch_sizerV   hidden_shapequery_statesr   r   attention_interfacequery_states_with_bias_uquery_states_with_bias_vrelative_key_states	matrix_bdr   r   s                      r&   r\   z ParakeetEncoderAttention.forward#  sj    $))#2.!,
J"JDMMB{{=166|DNNqRST[[/44\BLLQPQR
{{=166|DNNqRST(?(M(MKK,,.E)
 $0$++2B2Bt{{..4==3
 $
  $0$++2B2Bt{{..4==3
 $
  #223FG166z2t{{GfGfhlhuhuv -/B/J/J1aQRTU/VV	OOI.	c;J;./	,	% "..~/I/I/KUSY][I %8	%
*$#}}C$2H2HLL	%
 	%
!\ *k));;;;FFHkk+.L((r%   c                     |j                   \  }}}}t        j                  j                  |d      }|j	                  ||d|      }|ddddddf   j	                  ||||      }|S )ztRelative position shift for Shaw et al. style attention. See appendix B of https://huggingface.co/papers/1901.02860.)r   r   )padrB   Nr   )rK   r   rt   r   r   )r<   attention_scoresr   	num_headsquery_lengthposition_lengths         r&   r   z#ParakeetEncoderAttention._rel_shift\  st    ?O?U?U<
I|_==,,-=6,J+00YLY+Aq!"H5:::yR^`opr%   r]   )r   r   r    r^   r   intr4   r!   r"   r   r   tupler\   r   r`   ra   s   @r&   r   r     s    ~[4 [ [B /3	7)||7) #\\D07) t+	7)
 +,7) 
u||U\\)	*7)r r%   r   c                        e Zd Zdef fdZdej                  dej                  fdZ	d	dej                  dej                  fdZ
 xZS )
 ParakeetEncoderSubsamplingConv2Dr*   c                    t         |           |j                  | _        |j                  | _        |j                  | _        | j                  dz
  dz  | _        t        t        j                  |j                              | _        t        j                         | _        | j                   j#                  t        j$                  d| j                  | j                  | j
                  | j                               | j                   j#                  t        j&                                t)        | j                  dz
        D ]  }| j                   j#                  t        j$                  | j                  | j                  | j                  | j
                  | j                  | j                               | j                   j#                  t        j$                  | j                  | j                  d             | j                   j#                  t        j&                                 |j*                  | j
                  | j                  z  z  }t        j,                  |j                  |z  |j.                  d      | _        y )Nr   r.   )r{   r|   r}   )r{   r|   r}   r~   r{   Tre   )r3   r4   subsampling_conv_kernel_sizer{   subsampling_conv_strider|   subsampling_conv_channelsr   r}   r   mathlog2subsampling_factor
num_layersr   
ModuleListlayersappendConv2dReLUrangenum_mel_binsrg   r7   linear)r<   r*   i
out_lengthr>   s       r&   r4   z)ParakeetEncoderSubsamplingConv2D.__init__f  s   !>>4488((1,2dii(A(ABC mmoIIaD4D4DT[[bfbnbno	
 	2779%t*+ 	*AKK		MMMM $ 0 0;; LL==	 KKryySTUVKKrwwy)	*" ((T[[$//-IJ
ii @ @: MvOaOahlmr%   input_lengths
conv_layerc                     t        |d      rR|j                  dk7  rC|j                  }|j                  d   }|j                  d   }||d   z   |d   z   |z
  |z  dz   }|S |S )Nr|   )r   r   r   r   )hasattrr|   r}   r{   )r<   r   r   r}   r{   r|   output_lengthss          r&   _get_output_lengthz3ParakeetEncoderSubsamplingConv2D._get_output_length  sx    :x(Z->->&-H ((G$003K&&q)F+gaj871:ESX^^abbN!!r%   input_featuresr   c                    |j                  d      }||j                  d      nd }| j                  D ]  } ||      }t        |t        j
                        s&|)| j                  ||      }|j                  d   }t        j                  ||j                        |d d d f   k  }||d d d d d d f   z  } |j                  dd      j                  |j                  d   |j                  d   d      }| j                  |      }|S )Nr   rB   r.   rC   r   )r   sumr   rN   r   r   r  rK   r!   r6   r1   rQ   rU   r   )r<   r  r   r?   current_lengthslayercurrent_seq_lengthchannel_masks           r&   r\   z(ParakeetEncoderSubsamplingConv2D.forward  s   &0034B4N.,,R0TX[[ 
	@E!-0M %+0J"&"9"9/5"Q%2%8%8%;"LL!3N<Q<QRUdefhlelUmm  aq$.>!??
	@ &//15==m>Q>QRS>TVcViVijkVlnpqM2r%   r]   )r   r   r    r   r4   r!   r"   r   r   r  r\   r`   ra   s   @r&   r   r   e  sI    !n4 !nF	 	")) 	ell ELL r%   r   c                        e Zd Zddededz  f fdZ	 	 ddej                  dej                  dz  dej                  dz  dee	   d	ej                  f
d
Z
 xZS )ParakeetEncoderBlockNr*   r   c                    t         |           d| _        t        |      | _        t        ||      | _        t        |      | _        t        |      | _	        t        j                  |j                        | _        t        j                  |j                        | _        t        j                  |j                        | _        t        j                  |j                        | _        t        j                  |j                        | _        y NF)r3   r4   gradient_checkpointingrc   feed_forward1r   	self_attnrx   convfeed_forward2r   	LayerNormr7   norm_feed_forward1norm_self_att	norm_convnorm_feed_forward2norm_outr   s      r&   r4   zParakeetEncoderBlock.__init__  s    &+#7?1&)D4V<	7?"$,,v/A/A"B\\&*<*<=f&8&89"$,,v/A/A"BV%7%78r%   r?   r   r   r   r   c                 x   |}| j                  | j                  |            }|d|z  z   }| j                  |      } | j                  d|||d|\  }}||z   }| j	                  | j                  |      |      }	||	z   }| j                  | j                  |            }
|d|
z  z   }| j                  |      }|S )Ng      ?)r?   r   r   )r   r$   )	r  r  r  r  r  r  r  r  r  )r<   r?   r   r   r   residualnormalized_hidden_statesr   _conv_output
ff2_outputs              r&   r\   zParakeetEncoderBlock.forward  s     !**4+B+B=+QR 3#66#'#5#5m#D ' 
2) 3
 	
Q &3ii} =ni]%3''(?(?(NO
%j(88m4r%   r]   NN)r   r   r    r   r   r4   r!   r"   r   r   r\   r`   ra   s   @r&   r
  r
    sx    94 9t 9$ /337	|| t+ #\\D0	
 +, 
r%   r
  c                        e Zd ZU eed<   dZdZdZdZdgZ	dZ
dZdZdZdZdZeedZ ej(                          fd	       Zd
ej,                  fdZddej,                  dedz  fdZ xZS )ParakeetPreTrainedModelr*   modelr  audioTr
  F)r?   
attentionsc                    t         |   |       t        | j                  d      r| j                  j                  }n%t        | j                  j                         dd      }t        |t              rEt        j                  |j                  d|       t        j                  |j                  d|       y t        |t              ryddt        j                  d| j                  j                   dt        j"                  	      | j                  j                   z  z  z  }t        j$                  |j&                  |       y y )
Ninitializer_rangeg{Gz?r   )meanstdr-   r,   r   r.   r/   )r3   _init_weightsr   r*   r%  r   get_text_configrN   r   initnormal_r   r   r(   r!   r6   r7   r8   copy_r)   )r<   r   r'  r)   r>   s       r&   r(  z%ParakeetPreTrainedModel._init_weights  s    f%4;; 34++//C $++5579LdSCf67LLSc:LLSc: DEELLDKK,C,CQekkZ]a]h]h]t]ttuH JJv1	 Fr%   r   c                    t        | j                  t              r| j                  j                  n| j                  }|j                  }|j
                  }t        t        j                  |j                              }|dz
  dz  dz  }||z
  }|}t        |      D ]Q  }	t        j                  |j                  t        j                        |z   |      dz   }t        j                  |      }S |j                  t        j                        S )Nr   r.   r/   r-   )rN   r*   r   encoder_configr   r   r   r   r   r   r   r!   divr9   r:   floor)
r<   r   r.  r{   r|   r   all_paddingsadd_padlengthsr  s
             r&   _get_subsampling_output_lengthz6ParakeetPreTrainedModel._get_subsampling_output_length  s    7A$++O`7a33gkgrgr$AA77>#D#DEF
#aA-1,z" 	+Aii


 = GPSVVGkk'*G	+ zz		z**r%   Nr   target_lengthc                     | j                  |j                  d            }||n|j                         }t        j                  ||j
                        |dddf   k  }|S )z
        Convert the input attention mask to its subsampled form. `target_length` sets the desired output length, useful
        when the attention mask length differs from `sum(-1).max()` (i.e., when the longest sequence in the batch is padded)
        rB   NrC   )r4  r  maxr!   r6   r1   )r<   r   r5  r   
max_lengths        r&   _get_output_attention_maskz2ParakeetPreTrainedModel._get_output_attention_mask  sc    
 <<^=O=OPR=ST&3&?]^EWEWEY
j9N9NOR`abdhahRiir%   r]   )r   r   r    r   r#   base_model_prefixmain_input_nameinput_modalitiessupports_gradient_checkpointing_no_split_modules_supports_flat_attention_mask_supports_sdpa_supports_flex_attn_supports_flash_attn_can_compile_fullgraph_supports_attention_backendr
  r   _can_record_outputsr!   r_   r(  r"   r4  r   r9  r`   ra   s   @r&   r   r     s    &O&*#/0$(!N !!"&-.
 U]]_2 2&+ELL +"	 	VY\`V` 	r%   r   z{
    The Parakeet Encoder model, based on the [Fast Conformer architecture](https://huggingface.co/papers/2305.05084).
    c                        e Zd ZU eed<   dZdef fdZeee		 	 dde
j                  de
j                  dz  dedz  dee   d	ef
d
                     Z xZS )ParakeetEncoderr*   encoderc           	         t         |   |       || _        d| _        |j                  | _        |j
                  | _        |j                  | _        |j                  rt        j                  |j                        nd| _        t        |      | _        t        |      | _        t!        j"                  t%        |j&                        D cg c]  }t)        ||       c}      | _        | j-                          y c c}w )NFr-   )r3   r4   r*   r  ru   dropout_positions	layerdropscale_inputr   sqrtr7   input_scaler   subsamplingr(   encode_positionsr   r   r   num_hidden_layersr
  r   	post_initr   s      r&   r4   zParakeetEncoder.__init__)  s     &+#~~!'!9!9))<B<N<N499V%7%78TW;FC DV LmmFKFLdLdFef!&)4f
 	 gs   
C:Nr  r   output_attention_maskr   r   c                    | j                  ||      }|| j                  z  }| j                  |      }t        j                  j                  || j
                  | j                        }t        j                  j                  || j                  | j                        }|u| j                  ||j                  d         }|j                  d      j                  d|j                  d   d      }||j                  dd      z  }|j                  d      }| j                  D ]E  }d}	| j                  r&t        j                  g       }
|
| j                   k  rd}	|	r: ||f||d	|}G t#        ||rj%                         
      S d
      S )aJ  
        output_attention_mask (`bool`, *optional*):
            Whether to return the output attention mask.

        Example:

        ```python
        >>> from transformers import AutoProcessor, ParakeetEncoder
        >>> from datasets import load_dataset, Audio

        >>> model_id = "nvidia/parakeet-ctc-1.1b"
        >>> processor = AutoProcessor.from_pretrained(model_id)
        >>> encoder = ParakeetEncoder.from_pretrained(model_id)

        >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        >>> ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))

        >>> inputs = processor(ds[0]["audio"]["array"])
        >>> encoder_outputs = encoder(**inputs)

        >>> print(encoder_outputs.last_hidden_state.shape)
        ```
        rq   Nr   r5  rB   r.   FT)r   r   )last_hidden_stater   )rO  rN  rP  r   rt   ru   rs   rJ  r9  rK   r   rM   rQ   r   r!   randrK  r   r   )r<   r  r   rS  r   r?   r   output_maskencoder_layerto_dropdropout_probabilitys              r&   r\   zParakeetEncoder.forward<  s   D ((H%(8(88"33MB--mt||VZVcVc-d mm334#9#9DMM 4 
 %99.XeXkXklmXn9oK(2215<<RATATUVAWY[\N+n.F.Fq!.LLN+55a8N![[ 	MG}}&+jjn#&7"G -!!#1(;! 	!	  *+QfKOO<M
 	
lp
 	
r%   r  )r   r   r    r   r#   r:  r4   r   r   r   r!   r"   r   r   r   r   r\   r`   ra   s   @r&   rG  rG     s     "!!4 &  /3-1	@
@
 t+@
  $d{	@

 +,@
 
@
   @
r%   rG  c                       e Zd ZU dZej
                  ed<   dZeej                     dz  ed<   dZ
eeej                        dz  ed<   dZeeej                        dz  ed<   y)ParakeetGenerateOutputal  
    Outputs of Parakeet models.

    Args:
        sequences (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
            The generated sequences. The second dimension (sequence_length) is either equal to `max_length` or shorter
            if all batches finished early due to the `eos_token_id`.
        logits (`tuple(torch.FloatTensor)` *optional*, returned when `output_logits=True`):
            Unprocessed prediction scores of the language modeling head (scores for each vocabulary token before SoftMax)
            at each generation step. Tuple of `torch.FloatTensor` with up to `max_new_tokens` elements (one element for
            each generated token), with each tensor of shape `(batch_size, config.vocab_size)`.
        attentions (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `output_attentions=True`):
            Tuple (one element for each generated token) of tuples (one element for each layer of the decoder) of
            `torch.FloatTensor` of shape `(batch_size, num_heads, generated_length, sequence_length)`.
        hidden_states (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `output_hidden_states=True`):
            Tuple (one element for each generated token) of tuples (one element for each layer of the decoder) of
            `torch.FloatTensor` of shape `(batch_size, generated_length, hidden_size)`.
    	sequencesNlogitsr#  r?   )r   r   r    r^   r!   
LongTensorr#   r_  r   FloatTensorr#  r?   r$   r%   r&   r]  r]    sm    & .2FE%##$t+29=JeE--./$6=<@M5u0012T9@r%   r]  zS
    Parakeet Encoder with a Connectionist Temporal Classification (CTC) head.
    c                   L    e Zd ZU eed<   def fdZee	 	 ddej                  dej                  dz  dej                  dz  de
e   def
d	              Z ej                         	 	 ddej                  dej                  dz  d
ede
e   deej"                  z  f
d       Z xZS )ParakeetForCTCr*   c                     t         |   |       t        |j                        | _        t        j                  |j                  j                  |j                  d      | _	        | j                          y )Nr   r   )r3   r4   rG  r.  rH  r   r   r7   
vocab_sizectc_headrR  ro   s     r&   r4   zParakeetForCTC.__init__  sS     &v'<'<=		&"7"7"C"CVEVEVdefr%   Nr  r   labelsr   r   c           
          | j                   d||d|}|j                  }| j                  |j                  dd            j                  dd      }d}|Y||n$t	        j
                  |t        j                        }| j                  |j                  d            }	|| j                  j                  k7  }
|
j                  d      }|j                  |
      }t        j                  j                  |dt        j                        j                  dd      }t        j                   j"                  j%                  d	
      5  t        j                  j'                  |||	|| j                  j                  | j                  j(                  | j                  j*                        }ddd       t-        |||j.                  |j0                        S # 1 sw Y   ,xY w)a  
        Example:

        ```python
        >>> from transformers import AutoProcessor, ParakeetForCTC
        >>> from datasets import load_dataset, Audio

        >>> model_id = "nvidia/parakeet-ctc-1.1b"
        >>> processor = AutoProcessor.from_pretrained(model_id)
        >>> model = ParakeetForCTC.from_pretrained(model_id)

        >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        >>> ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))

        >>> inputs = processor(ds[0]["audio"]["array"], text=ds[0]["text"])
        >>> outputs = model(**inputs)

        >>> print(outputs.loss)
        ```r  r   r   r.   Nr/   rB   r   r   F)rG   )blank	reductionzero_infinity)lossr_  r?   r#  r$   )rH  rV  rf  rQ   r!   	ones_likelongr4  r  r*   pad_token_idmasked_selectr   rt   log_softmaxr   backendscudnnflagsctc_lossctc_loss_reductionctc_zero_infinityr   r?   r#  )r<   r  r   rg  r   encoder_outputsr?   r_  rm  r   labels_masktarget_lengthsflattened_targets	log_probss                 r&   r\   zParakeetForCTC.forward  s   : '$,, 
))
 
 (99}66q!<=GG1M #1"<%//R`hmhrhrBs  !??@R@RSU@VWM !DKK$<$<<K(__R0N & 4 4[ A 11&b1V``abdefI%%++E+: 	}}--%!"++22"kk<<"&++"?"? . 	 )77&11	
 	
	 	s   A#GGreturn_dict_in_generatec                 H   d|d<    | j                   d	||d|}|j                  j                  d      }|:| j                  ||j                  d         }| j
                  j                  || <   |r-t        ||j                  |j                  |j                        S |S )
a3  
        Example:

        ```python
        >>> from transformers import AutoProcessor, ParakeetForCTC
        >>> from datasets import load_dataset, Audio

        >>> model_id = "nvidia/parakeet-ctc-1.1b"
        >>> processor = AutoProcessor.from_pretrained(model_id)
        >>> model = ParakeetForCTC.from_pretrained(model_id)

        >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
        >>> ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))

        >>> inputs = processor(ds[0]["audio"]["array"], text=ds[0]["text"])
        >>> predicted_ids = model.generate(**inputs)
        >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)

        >>> print(transcription)
        ```
        Treturn_dictri  rB   rH   r   rU  )r^  r_  r#  r?   r$   )
r\   r_  argmaxr9  rK   r*   rp  r]  r#  r?   )r<   r  r   r~  r   outputsr^  s          r&   generatezParakeetForCTC.generate  s    : !%}".$,, #
))#
 #
 NN))b)1	 %!<<^[d[j[jkl[m<nN)-)A)AI~o&")#~~"--%33	  r%   r  r  )r   r   r    r   r#   r4   r   r   r!   r"   r   r   r   r\   r_   r   r]  r`  r  r`   ra   s   @r&   rc  rc    s    0   /3&*	E
E
 t+E
 t#	E

 +,E
 
E
  E
N U]]_ /3(-	33 t+3 "&	3
 +,3 
 %"2"2	23 3r%   rc  )rc  rG  r   )r   )r   )8r   collections.abcr   dataclassesr   r!   r    r   r*  activationsr   integrationsr	   r
   modeling_layersr   modeling_outputsr   r   modeling_utilsr   r   processing_utilsr   utilsr   r   r   r   utils.genericr   r   configuration_parakeetr   r   r   Moduler(   rc   rx   r   r   r"   r   r   r:   r   r   r   r
  r   rG  r]  rc  __all__r$   r%   r&   <module>r     s5  *  $ !   & ! I 9 ? F & V V ? L 
/ / //7299 /7d E-ryy E-P( *+ ,2	UU\\ 	U# 	U%,, 	U& %II%<<% 
% <<	%
 LL4'% % % '(%4 )*_ ryy _  +_ DBryy BJ,5 ,^ Co C CL 
Z
- Z

Z
z A[ A A4 
H, H
HV Kr%   