
    ^j]                     x   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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m Z  ddl!m"Z"m#Z# ddl$m%Z% ddl&m'Z'm(Z( ddl)m*Z*m+Z+m,Z, ddl-m.Z. ddl/m0Z0  ed       G d dejb                               Z2dejf                  de4dejf                  fdZ5	 d9dejb                  dejf                  dejf                  dejf                  d ejf                  dz  d!e6d"e6d#e%e*   fd$Z7 ed%      d:d&       Z8d' Z9 ee8       G d( d)ejb                               Z: G d* d+ejb                        Z; G d, d-e      Z< G d. d/ejb                        Z=e' G d0 d1e#             Z>e' G d2 d3e>             Z?e' G d4 d5e>e             Z@ G d6 d7ee>      ZAg d8ZBy);    )Callable)OptionalN   )initialization)ACT2FN)CacheDynamicCache)GenerationMixin)use_kernel_forward_from_hubuse_kernel_func_from_hubuse_kernelized_func)create_causal_mask!create_sliding_window_causal_mask) GenericForSequenceClassificationGradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)auto_docstringcan_return_tuple)TransformersKwargsmaybe_autocastmerge_with_config_defaults)capture_outputs   )Olmo3ConfigRMSNormc                   P     e Zd Zddeddf fdZdej                  fdZd Z xZ	S )Olmo3RMSNormepsreturnNc                     t         |           t        j                  t	        j
                  |            | _        || _        y)z;
        Olmo3RMSNorm is equivalent to T5LayerNorm
        N)super__init__nn	Parametertorchonesweightvariance_epsilon)selfhidden_sizer$   	__class__s      s/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/olmo3/modeling_olmo3.pyr(   zOlmo3RMSNorm.__init__.   s1     	ll5::k#:; #    c                 "   |j                   }|j                  t        j                        }|j	                  d      j                  dd      }|t        j                  || j                  z         z  }| j                  |z  j                  |      S )N   T)keepdim)	dtypetor+   float32powmeanrsqrtr.   r-   )r/   hidden_statesinput_dtypevariances       r2   forwardzOlmo3RMSNorm.forward6   sy    #))%((7 $$Q',,R,>%Ht?T?T4T(UUm+//<<r3   c                 ^    t        | j                  j                         d| j                   S )Nz, eps=)tupler-   shaper.   )r/   s    r2   
extra_reprzOlmo3RMSNorm.extra_repr=   s*    ))*+6$2G2G1HIIr3   )gư>)
__name__
__module____qualname__floatr(   r+   TensorrA   rE   __classcell__r1   s   @r2   r#   r#   ,   s,    $ $$ $= =Jr3   r#   r>   n_repr%   c                     | j                   \  }}}}|dk(  r| S | dddddddddf   j                  |||||      } | j                  |||z  ||      S )z
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    r   N)rD   expandreshape)r>   rM   batchnum_key_value_headsslenhead_dims         r2   	repeat_kvrU   A   so    
 2?1D1D.Ehz!!Qa"23::5BUW\^bdlmM  (;e(CT8TTr3   modulequerykeyvalueattention_maskscalingdropoutkwargsc                    t        || j                        }t        || j                        }	t        j                  ||j	                  dd            |z  }
||
|z   }
t
        j                  j                  |
dt        j                        j                  |j                        }
t
        j                  j                  |
|| j                        }
t        j                  |
|	      }|j	                  dd      j                         }||
fS )Nr5   r   r6   )dimr8   )ptrainingr   )rU   num_key_value_groupsr+   matmul	transposer)   
functionalsoftmaxr:   r9   r8   r\   ra   
contiguous)rV   rW   rX   rY   rZ   r[   r\   r]   
key_statesvalue_statesattn_weightsattn_outputs               r2   eager_attention_forwardrl   M   s     3 ; ;<JUF$?$?@L<<z';';Aq'ABWLL!#n4==((2U]](SVVW\WbWbcL==((6??([L,,|\:K''1-88:K$$r3   rotary_pos_embc                 
   | j                   |j                   }}|j                  |      }|j                  |      }| |z  t        |       |z  z   }||z  t        |      |z  z   }|j                  |      |j                  |      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.
    )r8   	unsqueezerotate_halfr9   )	qkcossinunsqueeze_dimq_typek_typeq_embedk_embeds	            r2   apply_rotary_pos_embrz   f   s|    & WWaggFF
--
&C
--
&C3w;q>C/0G3w;q>C/0G::fwzz&111r3   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..Nr6   r5   r_   )rD   r+   cat)xx1x2s      r2   rp   rp      sZ    	
3"!''"+"""	#B	
3q ""	#B99rc2YB''r3   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ej                  dz  d	e
dz  d
ee   de	ej                  ej                  dz  f   fdZ xZS )Olmo3Attentionz=Multi-headed attention from 'Attention Is All You Need' paperconfig	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                  z  |j*                        | _        t)        |j                  | j                  z  |j*                        | _        |j0                  |   | _        | j2                  dk(  r|j4                  | _        y d | _        y )NrT   g      Tbiassliding_attention)r'   r(   r   r   getattrr0   num_attention_headsrT   rR   rb   r[   attention_dropout	is_causalr)   Linearattention_biasq_projk_projv_projo_projr#   rms_norm_epsq_normk_normlayer_typesattention_typesliding_windowr/   r   r   r1   s      r2   r(   zOlmo3Attention.__init__   s   "
F4F4F&JdJd4de$*$>$>&B\B\$\!}}d*!'!9!9ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii : :T]] JQWQfQf
 ii&&68J8JQWQfQf
 #6#=#=#MvObObc"6#=#=#MvObObc$00;7;7J7JNa7af33gkr3   Nr>   position_embeddingsrZ   past_key_valuesr]   r%   c                 h   |j                   d d }g |d| j                  }| j                  | j                  |            }| j	                  | j                  |            }	| j                  |      }
|j                  |      j                  dd      }|	j                  |      j                  dd      }	|
j                  |      j                  dd      }
|\  }}t        ||	||      \  }}	| |j                  |	|
| j                        \  }	}
t        j                  | j                  j                  t               } || ||	|
|f| j"                  sdn| j$                  | j&                  | j(                  d|\  }} |j*                  g |d j-                         }| j/                  |      }||fS )Nr6   r   r5           )r\   r[   r   )rD   rT   r   r   r   r   r   viewrd   rz   updater   r   get_interfacer   _attn_implementationrl   ra   r   r[   r   rP   rg   r   )r/   r>   r   rZ   r   r]   input_shapehidden_shapequery_statesrh   ri   rs   rt   attention_interfacerk   rj   s                   r2   rA   zOlmo3Attention.forward   s    $))#2.88b8$--8{{4;;}#=>[[]!;<
{{=1#((6@@AF__\2<<QB
#((6@@AF&S#7jRUWZ#[ j&'6'='=j,X\XfXf'g$J(?(M(MKK,,.E)
 %8
%
  $}}C$2H2HLL..
%
 
%
!\ *k));;;;FFHkk+.L((r3   N)rF   rG   rH   __doc__r    intr(   r+   rJ   rC   r   r   r   rA   rK   rL   s   @r2   r   r      s    Gl{ ls l@ )-+)||+) #5<<#=>+) t+	+)
 +) +,+) 
u||U\\D00	1+)r3   r   c                   $     e Zd Z fdZd Z xZS )Olmo3MLPc                    t         |           || _        |j                  | _        |j                  | _        t        j                  | j                  | j                  d      | _        t        j                  | j                  | j                  d      | _        t        j                  | j                  | j                  d      | _	        t        |j                     | _        y NFr   )r'   r(   r   r0   intermediate_sizer)   r   	gate_projup_proj	down_projr   
hidden_actact_fnr/   r   r1   s     r2   r(   zOlmo3MLP.__init__   s    !--!'!9!94#3#3T5K5KRWXyy!1!143I3IPUV4#9#94;K;KRWXV../r3   c                     | j                  | j                  | j                  |            | j                  |      z        }|S r   )r   r   r   r   )r/   r~   r   s      r2   rA   zOlmo3MLP.forward   s6    NN4;;t~~a/@#ADLLQRO#ST	r3   )rF   rG   rH   r(   rA   rK   rL   s   @r2   r   r      s    0r3   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j                  dz  de	dz  d	e
dz  d
eej                  ej                  f   dz  dee   dej                  fdZ xZS )Olmo3DecoderLayerr   r   c                     t         |           |j                  | _        t        ||      | _        t        |      | _        t        |j                  |j                        | _	        t        |j                  |j                        | _
        y )N)r   r   r$   )r'   r(   r0   r   	self_attnr   mlpr#   r   post_attention_layernormpost_feedforward_layernormr   s      r2   r(   zOlmo3DecoderLayer.__init__   sl    !--'vKF#(4V5G5GVM`M`(a%*6v7I7IvObOb*c'r3   Nr>   rZ   position_idsr   	use_cacher   r]   r%   c           
          |} | j                   d||||||d|\  }}	| j                  |      }||z   }|}| j                  |      }| j                  |      }||z   }|S )N)r>   rZ   r   r   r   r    )r   r   r   r   )
r/   r>   rZ   r   r   r   r   r]   residual_s
             r2   rA   zOlmo3DecoderLayer.forward   s     !)4>> 
')%+ 3
 
q 55mD =0 !/77F =0r3   )NNNFN)rF   rG   rH   r    r   r(   r+   rJ   
LongTensorr   boolrC   r   r   rA   rK   rL   s   @r2   r   r      s    d{ ds d /304(,!&HL|| t+ &&-	
  $; #5<<#=>E +, 
r3   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z  d	ed
ef   f
d       Z ej                         edd              Z xZS )Olmo3RotaryEmbeddinginv_freqNr   c                 t   t         |           |j                  | _        |j                  | _        || _        t        t        |j                              | _        i | _	        | j                  D ]  }| j
                  j                  |   }||d   | j                  |<   | j                  }| j                  |   dk7  rt        | j                  |      } || j
                  |      \  }}| j                  | d|d       | j                  | d|j                         d       t        | | d|        y )	N	rope_typedefault
layer_type	_inv_freqF)
persistent_original_inv_freq_attention_scaling)r'   r(   max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenr   listsetr   r   rope_parameterscompute_default_rope_parametersr   register_bufferclonesetattr)	r/   r   devicer   rope_paramsrope_init_fncurr_inv_freqcurr_attention_scalingr1   s	           r2   r(   zOlmo3RotaryEmbedding.__init__  s6   "("@"@$*$B$B!F$6$6 78** 	UJ++55jAK")4[)ADNN:&%)%I%IL~~j)Y624>>*3MN4@Yc4d1M1  J<y!9=UZ [  J</A!BMDWDWDYfk lDZL(:;=ST	Ur3   r   ztorch.deviceseq_lenr   r%   z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.
            layer_type (`str`, *optional*):
                The current layer type if the model has different RoPE parameters per type.
                Should not be used unless `config.layer_types is not None`

        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_thetarT   Ng      ?r   r5   )r8   )r   r8   )	r   r   r0   r   r+   arangeint64r9   rI   )r   r   r   r   baser_   attention_factorr   s           r2   r   z4Olmo3RotaryEmbedding.compute_default_rope_parameters&  s    2 %%j1,?fj$/c63E3EIcIc3c U\\!S!5;;?BB&X]XcXcBdgjjk
 )))r3   c                    t        | | d      }t        | | d      }|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                         |z  }|
j                         |z  }d d d        ||fS # 1 sw Y   fS xY w)Nr   r   r   r6   r   mpscpuF)device_typeenabledr5   r|   )r   rI   rO   rD   r9   r   
isinstancetypestrr   rd   r+   r}   rs   rt   )r/   r~   r   r   r   attention_scalinginv_freq_expandedposition_ids_expandedr   freqsembrs   rt   s                r2   rA   zOlmo3RotaryEmbedding.forwardJ  sR    4J<y!9:#DZL8J*KL$T1d]399;BB<CUCUVWCXZ\^_`ccdedldlm ,QaZ 8 > > @'1!((--'E!((--[`J`ahhmmfkUC 	0&,,.1F1L1L1NNYYZ[]^_E))UEN3C'')//C'')//C		0 Cx	0 Cxs   *A1E''E3r   )NNNN)rF   rG   rH   r+   rJ   __annotations__r    r(   staticmethodr   r   r   rC   rI   r   no_gradr   rA   rK   rL   s   @r2   r   r     s    llU{ U* %)+/"!%	!*d"!*(!* t!* $J	!*
 
~u$	%!* !*F U]]_  r3   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 fdZ xZS )Olmo3PreTrainedModelr   modelTr   r   )r>   
attentionsc                    t         |   |       t        |t              r|j                  D ]  }|j
                  }|j                  |   dk7  rt        |j                  |      } ||j                  |      \  }}t        j                  t        || d      |       t        j                  t        || d      |        y y )Nr   r   r   r   )r'   _init_weightsr   r   r   r   r   r   r   initcopy_r   )r/   rV   r   r   r   r   r1   s         r2   r   z"Olmo3PreTrainedModel._init_weightsp  s    f%f23$00 ^
%EE##J/9<#6v7G7G
7S#TL#/*#U q

76j\+CDmT

76j\9K+LM}]^ 4r3   )rF   rG   rH   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   rK   rL   s   @r2   r   r   ^  s^    &*#,-#4"5N!"&*$
	^ 	^r3   r   c                        e Zd Zdef fdZeee	 	 	 	 	 	 ddej                  dz  dej                  dz  dej                  dz  dedz  dej                  dz  d	edz  d
ee   defd                     Z xZS )
Olmo3Modelr   c           	         t         |   |       |j                  | _        |j                  | _        t        j                  |j                  |j                  | j                        | _        t        j                  t        |j                        D cg c]  }t        ||       c}      | _        t        |j                  |j                        | _        t#        |      | _        d| _        | j)                          y c c}w )Nr   r   F)r'   r(   pad_token_idpadding_idx
vocab_sizer)   	Embeddingr0   embed_tokens
ModuleListrangenum_hidden_layersr   layersr#   r   normr   
rotary_embgradient_checkpointing	post_initr   s      r2   r(   zOlmo3Model.__init__~  s     !.. ++LL):):F<N<NPTP`P`ammCHIaIaCbcivy1c
 !!3!39L9LM	.f=&+# 	 ds   DN	input_idsrZ   r   r   inputs_embedsr   r]   r%   c           
      N   |d u |d uz  rt        d      || j                  |      }|r|t        | j                        }|V||j	                         nd}t        j                  |j                  d   |j                        |z   }|j                  d      }t        |x}	t              s)| j                  ||||d}
t        d
i |
t        d
i |
d}	|}i }t        | j                  j                        D ]  }| j!                  |||      ||<    t#        | j$                  d | j                  j&                         D ]G  \  }} ||f|	| j                  j                  |      |||| j                  j                  |      d|}I | j)                  |      }t+        ||	      S )Nz:You must specify exactly one of input_ids or inputs_embedsr  r   r   )r   )r   r  rZ   r   r   )full_attentionr   )rZ   r   r   r   )last_hidden_stater   r   )
ValueErrorr  r	   r   get_seq_lengthr+   r   rD   r   ro   r   dictr   r   r   r   r  	enumerater  r  r  r   )r/   r  rZ   r   r   r  r   r]   past_seen_tokenscausal_mask_mappingmask_kwargsr>   r   r   idecoder_layers                   r2   rA   zOlmo3Model.forward  s    -t";<YZZ *.*;*;I*FM0*$++>OCRC^==?de <<(;(;A(>}G[G[\_ooL'11!4L ?-F ++!."0#2 ,K #5"C{"C%F%U%U#
 & dkk556 	gJ.2oom\[e.f
+	g !*$++6U8U8U*V W 	A})24;;3J3J13MN) /$78O8OPQ8R$S M	 		-0&++
 	
r3   )NNNNNN)rF   rG   rH   r    r(   r   r   r   r+   r   rJ   r   FloatTensorr   r   r   r   rA   rK   rL   s   @r2   r
  r
  |  s    {     .2.204(,26!%;
##d*;
 t+;
 &&-	;

 ;
 ((4/;
 $;;
 +,;
 
!;
    ;
r3   r
  c                   B    e Zd ZddiZddiZddgdgfiZ 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e	j                  z  dee   defd              Z xZS )Olmo3ForCausalLMzlm_head.weightzmodel.embed_tokens.weightlm_headcolwise_gather_outputr>   logitsc                     t         |   |       t        |      | _        |j                  | _        t        j                  |j                  |j                  d      | _        | j                          y r   )
r'   r(   r
  r   r  r)   r   r0   r+  r  r   s     r2   r(   zOlmo3ForCausalLM.__init__  sU     '
 ++yy!3!3V5F5FUS 	r3   Nr  rZ   r   r   r  labelsr   logits_to_keepr]   r%   c	           
      x    | j                   d||||||d|	}
|
j                  }t        |t              rt	        | d      n|}| j                  |dd|ddf         }d}|* | j                  d||| j                  j                  d|	}t        |||
j                  |
j                  |
j                        S )a  
        Example:

        ```python
        >>> from transformers import AutoTokenizer, Olmo3ForCausalLM

        >>> model = Olmo3ForCausalLM.from_pretrained("meta-olmo3/Olmo3-2-7b-hf")
        >>> tokenizer = AutoTokenizer.from_pretrained("meta-olmo3/Olmo3-2-7b-hf")

        >>> 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."
        ```)r  rZ   r   r   r  r   N)r-  r/  r  )lossr-  r   r>   r   r   )r   r  r   r   slicer+  loss_functionr   r  r   r   r>   r   )r/   r  rZ   r   r   r  r/  r   r0  r]   outputsr>   slice_indicesr-  r2  s                  r2   rA   zOlmo3ForCausalLM.forward  s    > ,64:: ,
)%+',
 ,
  118B>SV8W~ot4]kmA}a,?@A%4%%pVFt{{OeOepiopD%#33!//))
 	
r3   )NNNNNNNr   )rF   rG   rH   _tied_weights_keys_tp_plan_pp_planr(   r   r   r+   r   rJ   r   r(  r   r   r   r   r   rA   rK   rL   s   @r2   r*  r*    s   *,GH23H_-z:;H  .2.204(,26*.!%-.6
##d*6
 t+6
 &&-	6

 6
 ((4/6
   4'6
 $;6
 ell*6
 +,6
 
 6
  6
r3   r*  c                       e Zd Zy)Olmo3ForSequenceClassificationN)rF   rG   rH   r   r3   r2   r;  r;    s    r3   r;  )r*  r;  r
  r   )r   )r   )Ccollections.abcr   typingr   r+   torch.nnr)    r   r   activationsr   cache_utilsr   r	   
generationr
   integrationsr   r   r   masking_utilsr   r   modeling_layersr   r   modeling_outputsr   r   modeling_rope_utilsr   r   modeling_utilsr   r   processing_utilsr   utilsr   r   utils.genericr   r   r   utils.output_capturingr   configuration_olmo3r    Moduler#   rJ   r   rU   rI   rl   rz   rp   r   r   r   r   r   r
  r*  r;  __all__r   r3   r2   <module>rP     s  * %    & ! . ) f f R [ O K F & 5 [ [ 5 , Y'J299 J (J(	UU\\ 	U# 	U%,, 	U& %II%<<% 
% <<	%
 LL4'% % % '(%2 *+2 ,24( )*I)RYY I) +I)Xryy  &2 &RM299 M` ^? ^ ^: O
% O
 O
d F
+_ F
 F
R	%EG[ 	 gr3   