
    i>~                     @   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 dd	lmZmZ dd
lmZ ddlmZmZmZ ddl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"m#Z# ddl$m%Z% ddl&m'Z'm(Z( ddl)m*Z*m+Z+m,Z, ddl-m.Z.  ed       G d dej^                               Z0 G d dej^                        Z1 G d dej^                        Z2 G d dej^                        Z3 G d d ej^                        Z4d! Z5 ed"      d?d#       Z6d$ejn                  d%e8d&ejn                  fd'Z9	 d@d(ej^                  d)ejn                  d*ejn                  d+ejn                  d,ejn                  dz  d-e:d.e:d/e%e'   fd0Z; ee6       G d1 d2ej^                               Z< G d3 d4e      Z=e( G d5 d6e#             Z>e( G d7 d8e>             Z?	 	 	 dAd9ejn                  e@ejn                     z  dz  d:e8dz  d,ejn                  dz  d&ejn                  e8z  fd;ZAe( G d< d=e>e             ZBg d>ZCy)B    )Callable)OptionalN)nn)
functional   )initialization)ACT2FN)CacheDynamicCache)GenerationMixin)use_kernel_forward_from_hubuse_kernel_func_from_hubuse_kernelized_func)create_causal_mask)GradientCheckpointingLayer)MoeCausalLMOutputWithPastMoeModelOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstring)can_return_tuplecheck_model_inputsmaybe_autocast   )GraniteMoeConfigRMSNormc                   ,     e Zd Zd fd	Zd Zd Z xZS )GraniteMoeRMSNormc                     t         |           t        j                  t	        j
                  |            | _        || _        y)z@
        GraniteMoeRMSNorm is equivalent to T5LayerNorm
        N)super__init__r   	Parametertorchonesweightvariance_epsilon)selfhidden_sizeeps	__class__s      |/home/obispo/Crisostomo_bridge/mision_env/lib/python3.12/site-packages/transformers/models/granitemoe/modeling_granitemoe.pyr%   zGraniteMoeRMSNorm.__init__/   s1     	ll5::k#:; #    c                 "   |j                   }|j                  t        j                        }|j	                  d      j                  dd      }|t        j                  || j                  z         z  }| j                  |j                  |      z  S )N   T)keepdim)	dtypetor'   float32powmeanrsqrtr*   r)   )r+   hidden_statesinput_dtypevariances       r/   forwardzGraniteMoeRMSNorm.forward7   sy    #))%((7 $$Q',,R,>%Ht?T?T4T(UU{{]--k:::r0   c                 ^    t        | j                  j                         d| j                   S )Nz, eps=)tupler)   shaper*   )r+   s    r/   
extra_reprzGraniteMoeRMSNorm.extra_repr>   s*    ))*+6$2G2G1HIIr0   )gư>)__name__
__module____qualname__r%   r>   rB   __classcell__r.   s   @r/   r"   r"   -   s    $;Jr0   r"   c                        e Zd ZU ej                  ed<   ddef fdZe	 	 	 ddedz  de	d   de
dz  ded	ef   fd
       Z ej                         ed               Z xZS )GraniteMoeRotaryEmbeddinginv_freqNconfigc                    t         |           |j                  | _        |j                  | _        || _        | j
                  j                  d   | _        | j                  }| j                  dk7  rt        | j                     } || j
                  |      \  }| _
        | j                  d|d       | j                  d|j                         d       y )N	rope_typedefaultrJ   F)
persistentoriginal_inv_freq)r$   r%   max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrK   rope_parametersrM   compute_default_rope_parametersr   attention_scalingregister_bufferclone)r+   rK   devicerope_init_fnrJ   r.   s        r/   r%   z"GraniteMoeRotaryEmbedding.__init__E   s    "("@"@$*$B$B!44[A!%!E!E>>Y&.t~~>L+7V+L($(ZeD0(..2BuUr0   rY   ztorch.deviceseq_lenreturnztorch.Tensorc                    | j                   d   }t        | dd      xs | j                  | j                  z  }d}d|t	        j
                  d|dt        j                        j                  |t        j                        |z  z  z  }||fS )	a  
        Computes the inverse frequencies according to the original RoPE implementation
        Args:
            config ([`~transformers.PreTrainedConfig`]):
                The model configuration.
            device (`torch.device`):
                The device to use for initialization of the inverse frequencies.
            seq_len (`int`, *optional*):
                The current sequence length. Unused for this type of RoPE.
        Returns:
            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
        
rope_thetahead_dimNg      ?r   r2   r5   )rY   r5   )	rT   getattrr,   num_attention_headsr'   arangeint64r6   float)rK   rY   r[   basedimattention_factorrJ   s          r/   rU   z9GraniteMoeRotaryEmbedding.compute_default_rope_parametersU   s    & %%l3fj$/c63E3EIcIc3c U\\!S!5;;?BB&X]XcXcBdgjjk
 )))r0   c                 N   | j                   d d d d f   j                         j                  |j                  d   dd      j	                  |j
                        }|d d d d d f   j                         }t        |j
                  j                  t              r/|j
                  j                  dk7  r|j
                  j                  nd}t        |d      5  |j                         |j                         z  j                  dd      }t        j                  ||fd	      }|j                         | j                  z  }|j                         | j                  z  }	d d d        j	                  |j                   
      	j	                  |j                   
      fS # 1 sw Y   AxY w)Nr   r3   r   mpscpuF)device_typeenabledr2   rg   r`   )rJ   re   expandrA   r6   rY   
isinstancetypestrr   	transposer'   catcosrV   sinr5   )
r+   xposition_idsinv_freq_expandedposition_ids_expandedrl   freqsembru   rv   s
             r/   r>   z!GraniteMoeRotaryEmbedding.forwards   sR    !MM$4-8>>@GGHZHZ[\H]_acdehhijiqiqr ,QaZ 8 > > @'1!((--'E!((--[`J`ahhmmfkUC 	5&,,.1F1L1L1NNYYZ[]^_E))UEN3C'')d444C'')d444C		5 vvAGGv$cff177f&;;;	5 	5s   BFF$)N)NNN)rC   rD   rE   r'   Tensor__annotations__r   r%   staticmethodr   intr@   re   rU   no_gradr   r>   rF   rG   s   @r/   rI   rI   B   s    llV/ V  *.+/"* 4'*(* t* 
~u$	%	* *: U]]_<  <r0   rI   c                   6     e Zd Zdedededdf fdZd Z xZS )GraniteMoeParallelExpertsnum_experts
input_sizeoutput_sizer\   Nc                     t         |           t        j                  t	        j
                  |||            | _        || _        || _        || _	        y)a  
        Initialize the GraniteMoeParallelExperts module.
        The experts weights are stored in [num_experts, output_size, input_size] format. Such that it's compatible with
        many MoE libraries, such as [Megablock](https://github.com/databricks/megablocks) and
        [ScatterMoE](https://github.com/shawntan/scattermoe), as well as the
        [MoE kernel](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/fused_moe/fused_moe.py)
        used in vllm.

        Args:
            num_experts (int):
                Number of experts.
            input_size (int):
                Size of the input.
            output_size (int):
                Size of the output.
        N)
r$   r%   r   r&   r'   emptyr)   r   r   r   )r+   r   r   r   r.   s       r/   r%   z"GraniteMoeParallelExperts.__init__   sD    " 	ll5;;{K#TU&$&r0   c                     |j                  |d      }g }t        | j                        D ]7  }|j                  t	        j
                  ||   | j                  |                9 t        j                  |d      }|S )a  
        Forward pass of the GraniteMoeParallelExperts module.

        Args:
            inputs (Tensor):
                Input tensor.
            expert_size:
                Expert size information.

        Returns:
            Tensor: Output tensor.
        r   rn   )	splitranger   appendFlinearr)   r'   rt   )r+   inputsexpert_size
input_listoutput_listiresultss          r/   r>   z!GraniteMoeParallelExperts.forward   sq     \\+1\5
t''( 	HAqxx
1t{{1~FG	H))KQ/r0   rC   rD   rE   r   r%   r>   rF   rG   s   @r/   r   r      s)    'C 'S 's 't '.r0   r   c                   2     e Zd Zdededef fdZd Z xZS )GraniteMoeTopKGatingr   r   top_kc                     t         |           || _        || _        || _        t        j                  ||d      | _        y)a  
        Initialize the top-k gating mechanism.

        Args:
            input_size (`int`):
                Size of the input.
            num_experts (`int`):
                Number of experts.
            top_k (`int`):
                Number of top experts to select.
        FbiasN)r$   r%   r   r   r   r   Linearlayer)r+   r   r   r   r.   s       r/   r%   zGraniteMoeTopKGating.__init__   s:     	&$
YYz;UC
r0   c                    | j                  |      j                         }|j                  | j                  d      \  }}t	        j
                  |d      j                  |      }t	        j                  |j                  d      | j                  g|j                  |j                        }|j                  d|d      }|j                         j                  d      }|j                         }|j!                         }	|	j#                  d      \  }
}|j%                  | j                  d      }|j!                         }||   }|||||fS )Nr   rn   r   r5   rY   trunc)rounding_mode)r   re   topkr   r'   softmaxtype_aszerossizer   r5   rY   scatterlongsumtolistflattensortdiv)r+   r;   logitstop_k_logitstop_k_indicestop_k_gatesr   gatesr   top_k_experts_index_sorted_expertsbatch_indexbatch_gatess                 r/   r>   zGraniteMoeTopKGating.forward   s.   M*002&,kk$**!k&D#mmmLa8@@O a $"2"23;;L;LU`UgUg
 a2jjl&&q) "((* &--/"/"4"4Q"7*..tzz.Q "))+!"67#[+{FRRr0   r   rG   s   @r/   r   r      s'    D3 DS D D(Sr0   r   c                   .     e Zd ZdZdef fdZd Z xZS )GraniteMoeMoEz
    A Sparsely gated mixture of experts layer with 1-layer Feed-Forward networks as experts.

    Args:
        config:
            Configuration object with model hyperparameters.
    rK   c                    t         |           |j                  | _        |j                  | _        t
        |j                     | _        t        |j                  | j                  | j                  dz        | _
        t        |j                  | j                  | j                        | _        t        | j                  |j                  |j                        | _        y )Nr2   )r   r   r   )r$   r%   r,   r   intermediate_sizer	   
hidden_act
activationr   num_local_expertsinput_linearoutput_linearr   num_experts_per_tokrouterr+   rK   r.   s     r/   r%   zGraniteMoeMoE.__init__   s     ,,!33 !2!235f6N6NPTP_P_aeaqaqtuauv6v7O7OQUQaQacgcrcrs*00,,
r0   c                    |j                         \  }}}|j                  d|      }| j                  |      \  }}}}}||   }	| j                  |	|      }
|
j	                  dd      }| j                  |d         |d   z  }
| j                  |
|      }||d d d f   z  }t        j                  ||z  | j                  f|j                  |j                        }|j                  d||      }|j                  ||| j                        }|S )Nr3   r2   rn   r   r   r   )r   reshaper   r   chunkr   r   r'   r   r   r5   rY   	index_addview)r+   layer_inputbszlengthemb_sizer   r   r   r   expert_inputsr;   chunked_hidden_statesexpert_outputsr   layer_outputs                  r/   r>   zGraniteMoeMoE.forward   s    + 0 0 2VX!))"h76:kk+6N3;[!#K0))-E - 3 3A2 3 >(=a(@ADYZ[D\\++M;G'+ag*>>S6\4??;>CWCW`n`u`uvq+~F#((fdooFr0   )rC   rD   rE   __doc__r   r%   r>   rF   rG   s   @r/   r   r      s    
/ 
r0   r   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..Nr3   r2   rn   )rA   r'   rt   )rw   x1x2s      r/   rotate_halfr     sZ    	
3"!''"+"""	#B	
3q ""	#B99rc2YB''r0   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kru   rv   unsqueeze_dimq_embedk_embeds          r/   apply_rotary_pos_embr     sY    & --
&C
--
&C3w;q>C/0G3w;q>C/0GGr0   r;   n_repr\   c                     | 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)rA   ro   r   )r;   r   batchnum_key_value_headsslenr_   s         r/   	repeat_kvr   -  so    
 2?1D1D.Ehz!!Qa"23::5BUW\^bdlmM  (;e(CT8TTr0   modulequerykeyvalueattention_maskscalingdropout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 )Nr2   r   r3   )rg   r5   )ptrainingr   )r   num_key_value_groupsr'   matmulrs   rA   r   r   r   r7   r6   r5   r   r   
contiguous)r   r   r   r   r   r   r   r   
key_statesvalue_statesattn_weightscausal_maskattn_outputs                r/   eager_attention_forwardr   9  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$$r0   c                       e Zd ZdZdedef fdZ	 	 	 	 ddej                  de	ej                  ej                  f   dz  dej                  dz  d	e
dz  d
ej                  dz  dee   de	ej                  ej                  f   fdZ xZS )GraniteMoeAttentionz=Multi-headed attention from 'Attention Is All You Need' paperrK   	layer_idxc                 ^   t         |           || _        || _        t	        |d|j
                  |j                  z        | _        |j                  |j                  z  | _	        |j                  | _        |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                         | _        y )Nr_   Tr   )r$   r%   rK   r   ra   r,   rb   r_   r   r   attention_multiplierr   attention_dropout	is_causalr   r   attention_biasq_projk_projv_projo_projr+   rK   r   r.   s      r/   r%   zGraniteMoeAttention.__init__W  sJ   "
F4F4F&JdJd4de$*$>$>&B\B\$\!22!'!9!9ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii&&68J8JQWQfQf
r0   Nr;   position_embeddingsr   past_key_valuescache_positionr   r\   c                    |j                   d d }g |d| j                  }| j                  |      j                  |      j	                  dd      }	| j                  |      j                  |      j	                  dd      }
| j                  |      j                  |      j	                  dd      }|\  }}t        |	|
||      \  }	}
|'|||d}|j                  |
|| j                  |      \  }
}t        j                  | j                  j                  t              } || |	|
||f| j                  sdn| j                   | j"                  d|\  }} |j$                  g |d j'                         }| j)                  |      }||fS )Nr3   r   r2   )rv   ru   r	          )r   r   )rA   r_   r  r   rs   r  r  r   updater   r   get_interfacerK   _attn_implementationr   r   r   r   r   r   r  )r+   r;   r  r   r  r	  r   input_shapehidden_shapequery_statesr   r   ru   rv   cache_kwargsattention_interfacer   r   s                     r/   r>   zGraniteMoeAttention.forwardn  s    $))#2.88b8$--8{{=166|DNNqRST[[/44\BLLQPQR
{{=166|DNNqRST&S#7jRUWZ#[ j&#&snUL'6'='=j,X\XfXfht'u$J(?(M(MKK,,.E)
 %8	%
  $}}C$2H2HLL	%
 	%
!\ *k));;;;FFHkk+.L((r0   NNNN)rC   rD   rE   r   r   r   r%   r'   r}   r@   r
   
LongTensorr   r   r>   rF   rG   s   @r/   r   r   S  s    G
/ 
C 
4 IM.2(,26))||)) #5<<#=>E)) t+	))
 )) ((4/)) +,)) 
u||U\\)	*))r0   r   c                        e Zd Zdedef fdZ	 	 	 	 ddej                  dej                  dz  dedz  dej                  dz  d	e
ej                  ej                  f   dz  d
ej                  fdZ xZS )GraniteMoeDecoderLayerrK   r   c                 B   t         |           |j                  | _        t        ||      | _        t        |j                  |j                        | _        t        |j                  |j                        | _        t        |      | _
        |j                  | _        y )N)rK   r   r-   )r$   r%   r,   r   	self_attnr"   rms_norm_epsinput_layernormpost_attention_layernormr   block_sparse_moeresidual_multiplierr  s      r/   r%   zGraniteMoeDecoderLayer.__init__  s|    !--,FiP01C1CI\I\](9&:L:LRXReRe(f% -f 5#)#=#= r0   Nr;   r   r  r	  r  r\   c           	          |}| j                  |      } | j                  d|||||d|\  }}||| j                  z  z   }|}| j                  |      }| j	                  |      }||| j                  z  z   }|S )N)r;   r   r  r	  r   )r  r  r  r  r  )	r+   r;   r   r  r	  r  r   residualr   s	            r/   r>   zGraniteMoeDecoderLayer.forward  s     !,,];)4>> 
')+) 3
 
q !=43K3K#KK 55mD--m< =43K3K#KKr0   r  )rC   rD   rE   r   r   r%   r'   r}   r
   r  r@   r>   rF   rG   s   @r/   r  r    s    >/ >C > /3(,26HL|| t+ 	
 ((4/ #5<<#=>E 
r0   r  c                        e Zd ZU eed<   dZdZdgZdgZdZ	dZ
dZdZdZeedZ ej$                          fd       Z xZS )	GraniteMoePreTrainedModelrK   modelTr  r  F)r;   
attentionsc                     t         |   |       t        |t              r7t	        j
                  |j                  d| j                  j                         y y )Nr  )r9   std)	r$   _init_weightsrp   r   initnormal_r)   rK   initializer_range)r+   r   r.   s     r/   r)  z'GraniteMoePreTrainedModel._init_weights  s>    f%f78LLSdkk6S6ST 9r0   )rC   rD   rE   r   r~   base_model_prefixsupports_gradient_checkpointing_no_split_modules_skip_keys_device_placement_supports_flash_attn_supports_sdpa_supports_flex_attn_can_compile_fullgraph_supports_attention_backendr  r   _can_record_outputsr'   r   r)  rF   rG   s   @r/   r$  r$    sp    &*#12#4"5N""&/)
 U]]_U Ur0   r$  c                       e Zd Zdef fdZee	 	 	 	 	 	 	 ddej                  dz  dej                  dz  dej                  dz  de
dz  dej                  dz  d	edz  d
ej                  dz  dee   defd              Z xZS )GraniteMoeModelrK   c           	      .   t         |   |       |j                  | _        |j                  | _        t        j                  |j                  |j                  | j                        | _        t        j                  t        |j                        D cg c]  }t        ||       c}      | _        t        |j                  |j                        | _        t#        |      | _        d| _        |j(                  | _        | j+                          y c c}w )Nr  rK   F)r$   r%   pad_token_idpadding_idx
vocab_sizer   	Embeddingr,   embed_tokens
ModuleListr   num_hidden_layersr  layersr"   r  normrI   
rotary_embgradient_checkpointingembedding_multiplier	post_initr  s      r/   r%   zGraniteMoeModel.__init__  s     !.. ++LL):):F<N<NPTP`P`ammHMfNfNfHgh9#FI6h
 &f&8&8f>Q>QR	36B&+#$*$?$?! 	 is   DN	input_idsr   rx   r  inputs_embeds	use_cacher	  r   r\   c                 b   |d u |d uz  rt        d      |r|t        | j                        }|| j                  |      }|F||j	                         nd}	t        j                  |	|	|j                  d   z   |j                        }||j                  d      }t        | j                  |||||      }
|| j                  z  }|}| j                  ||      }| j                  d | j                  j                   D ]  } ||f||
||||d|} | j                  |      }t!        ||      S )	Nz:You must specify exactly one of input_ids or inputs_embedsr:  r   r   )rY   )rK   input_embedsr   r	  r  rx   )r  r   rx   r  rJ  r	  )last_hidden_stater  )
ValueErrorr   rK   r?  get_seq_lengthr'   rc   rA   rY   r   r   rF  rD  rB  rA  rC  r   )r+   rH  r   rx   r  rI  rJ  r	  r   past_seen_tokensr   r;   r  decoder_layers                 r/   r>   zGraniteMoeModel.forward  sp    -t";<YZZ0*$++>O  --i8M!CRC^==?de"\\ "2]5H5H5K"KTaThThN )33A6L(;;&))+%
 &(A(AA% #oom\J![[)H4;;+H+HI 
	M)	$7*) /#-	 	M
	 		-0%++
 	
r0   )NNNNNNN)rC   rD   rE   r   r%   r   r   r'   r  r}   r
   FloatTensorboolr   r   r   r>   rF   rG   s   @r/   r8  r8    s    / "  .2.204(,26!%26;
##d*;
 t+;
 &&-	;

 ;
 ((4/;
 $;;
 ((4/;
 +,;
 
 ;
  ;
r0   r8  gate_logitsr   c                    | t        | t              syt        | t              rC| d   j                  }t        j                  | D cg c]  }|j                  |       c}d      }t        j                  j                  j                  d      }t        j                  ||d      \  }}	t        j                  j                  j                  |	|      }
|>t        j                  |
j                         d      }t        j                  |d      }n|j                  \  }}|j                  d   ||z  z  }|dddddddf   j                  |||||f      j                  d||      j                        }t        j                   |
j                         |z  d      t        j                   |d      z  }|ddddddf   j                  ||||f      j                  d|      j                  |      }t        j                   ||z  d      t        j                   |d      z  }t        j                   ||j#                  d      z        }||z  S c c}w )a  
    Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.

    See Switch Transformer (https://huggingface.co/papers/2101.03961) for more details. This function implements the loss
    function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between
    experts is too unbalanced.

    Args:
        gate_logits:
            Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of
            shape [batch_size X sequence_length, num_experts].
        num_experts:
            Number of experts
        top_k:
            The number of experts to route per-token, can be also interpreted as the `top-k` routing
            parameter.
        attention_mask (`torch.Tensor`, *optional*):
            The attention_mask used in forward function
            shape [batch_size X sequence_length] if not None.

    Returns:
        The auxiliary loss.
    Nr   rn   r3   )rp   r@   rY   r'   rt   r6   r   r   r   r   one_hotr9   re   rA   ro   r   r   r   )rT  r   r   r   compute_device
layer_gateconcatenated_gate_logitsrouting_weightsr   selected_expertsexpert_masktokens_per_expertrouter_prob_per_expert
batch_sizesequence_lengthrA  expert_attention_mask router_per_expert_attention_maskoverall_losss                      r/   load_balancing_loss_funcrd  *  s9   : *[%"@+u%$Q..#(99^i-jPZjmmN.K-jpq#r hh))112JPR1SO**_eDA((%%--.>LK!JJ{'8'8':B "'O!C&4&:&:#
O4::1=*B^_ 4AtT12V&
OUKXYWR,R	 	 "IIk&7&7&9<Q&QWXY\a\e\e!q]
 
 4At+,V&
O[QRWR%R	 	) "'?=]+]cd!ehmhqhq,!i
 "
 99.1G1Q1QRS1TTUL+%%[ .ks   Ic                   d    e Zd ZddiZddiZddgdgfiZdef fdZee		 	 	 	 	 	 	 	 	 dd
e
j                  d	z  de
j                  d	z  de
j                  d	z  ded	z  de
j                  d	z  de
j                  d	z  ded	z  de
j                  d	z  dee
j                  z  deez  fd              Z xZS )GraniteMoeForCausalLMzlm_head.weightzmodel.embed_tokens.weightlm_headcolwise_gather_outputr;   r   rK   c                 p   t         |   |       t        |      | _        |j                  | _        t        j                  |j                  |j                  d      | _        |j                  | _	        |j                  | _        |j                  | _        |j                  | _        | j                          y )NFr   )r$   r%   r8  r%  r=  r   r   r,   rg  router_aux_loss_coefr   r   r   logits_scalingrG  r   s     r/   r%   zGraniteMoeForCausalLM.__init__  s     $V,
 ++yy!3!3V5F5FUS$*$?$?!!33#)#=#= $33 	r0   NrH  r   rx   r  rI  labelsoutput_router_logitsr	  logits_to_keepr\   c
           
         ||n| j                   j                  } | j                  d||||||d|
}|j                  }t	        |	t
              rt        |	 d      n|	}| j                  |dd|ddf         }|| j                   j                  z  }d}|* | j                  ||fd| j                   j                  i|
}d}|rYt        |j                  | j                  | j                  |      }|+|| j                  |j!                  |j"                        z  z  }t%        ||||j&                  |j(                  |j*                  |j                        S )al  
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.

        Example:

        ```python
        >>> from transformers import AutoTokenizer, GraniteMoeForCausalLM

        >>> model = GraniteMoeForCausalLM.from_pretrained("ibm/PowerMoE-3b")
        >>> tokenizer = AutoTokenizer.from_pretrained("ibm/PowerMoE-3b")

        >>> prompt = "Hey, are you conscious? Can you talk to me?"
        >>> inputs = tokenizer(prompt, return_tensors="pt")

        >>> # Generate
        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
        ```N)rH  r   rx   r  rI  r	  r=  )lossaux_lossr   r  r;   r&  router_logitsr!  )rK   rm  r%  rM  rp   r   slicerg  rk  loss_functionr=  rd  rr  r   r   rj  r6   rY   r   r  r;   r&  )r+   rH  r   rx   r  rI  rl  rm  r	  rn  r   outputsr;   slice_indicesr   rp  rq  s                    r/   r>   zGraniteMoeForCausalLM.forward  s   L %9$D $++JjJj 	 $** 
)%+')
 
  118B>SV8W~ot4]kmA}a,?@A$++444%4%%  ;;11 	D /%%  ((	H !11HKK4LLL(#33!//))!//
 	
r0   )	NNNNNNNNr   )rC   rD   rE   _tied_weights_keys_tp_plan_pp_planr   r%   r   r   r'   r  r}   r
   rR  rS  r   r@   r   r>   rF   rG   s   @r/   rf  rf  |  s9   *,GH23H_-z:;H/   .2.204(,26*.,026-.S
##d*S
 t+S
 &&-	S

 S
 ((4/S
   4'S
 #TkS
 ((4/S
 ell*S
 
*	*S
  S
r0   rf  )rf  r8  r$  )r   )r  )Nr2   N)Dcollections.abcr   typingr   r'   r   torch.nnr   r    r   r*  activationsr	   cache_utilsr
   r   
generationr   integrationsr   r   r   masking_utilsr   modeling_layersr   modeling_outputsr   r   modeling_rope_utilsr   r   modeling_utilsr   r   processing_utilsr   utilsr   r   utils.genericr   r   r   configuration_granitemoer   Moduler"   rI   r   r   r   r   r   r}   r   r   re   r   r   r  r$  r8  r@   rd  rf  __all__r!  r0   r/   <module>r     ss  , %    $ & ! . ) f f / 9 Q K F & 7 Q Q 6 Y'J		 J (J(><		 ><B*		 *Z.S299 .Sb(BII (V( *+ ,2	UU\\ 	U# 	U%,, 	U& %II%<<% 
% <<	%
 LL4'% % % '(%4 )*C)")) C) +C)L"7 "J U U U. O
/ O
 O
h #
*.	O&ell 33d:O&tO& LL4'	O&
 \\CO&d g
5 g
 g
T Tr0   