a
    h[O                  
   @   s  d dl mZmZ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 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$m%Z% ddl&m'Z' ddl(m)Z) edG dd dej*Z+ej,e-ej,dddZ.d1ej*ej,ej,ej,eej, e/e/e e dddZ0d2ddZ1d d! Z2G d"d# d#ej*Z3G d$d% d%ej*Z4G d&d' d'eZ5G d(d) d)ej*Z6e"G d*d+ d+eZ7e"G d,d- d-e7Z8e"G d.d/ d/e7eZ9g d0Z:dS )3    )CallableOptionalUnionN)TransformersKwargs   )ACT2FN)CacheDynamicCache)GenerationMixin)use_kernel_forward_from_hub)create_causal_mask)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)auto_docstringcan_return_tuple)deprecate_kwarg)check_model_inputs   )Olmo2ConfigZRMSNormc                       s.   e Zd Zd fdd	Zdd Zdd Z  ZS )	Olmo2RMSNormư>c                    s&   t    tt|| _|| _dS )z;
        Olmo2RMSNorm is equivalent to T5LayerNorm
        N)super__init__nn	ParametertorchZonesweightvariance_epsilon)selfhidden_sizeeps	__class__ d/var/www/html/assistant/venv/lib/python3.9/site-packages/transformers/models/olmo2/modeling_olmo2.pyr       s    
zOlmo2RMSNorm.__init__c                 C   sJ   |j }|tj}|djddd}|t|| j  }| j| |S )N   T)Zkeepdim)	dtypetor!   float32powmeanZrsqrtr#   r"   )r$   hidden_statesZinput_dtypeZvariancer)   r)   r*   forward(   s
    zOlmo2RMSNorm.forwardc                 C   s   t | jj d| j S )Nz, eps=)tupler"   shaper#   )r$   r)   r)   r*   
extra_repr/   s    zOlmo2RMSNorm.extra_repr)r   )__name__
__module____qualname__r   r3   r6   __classcell__r)   r)   r'   r*   r      s   r   )r2   n_repreturnc                 C   s^   | j \}}}}|dkr| S | dddddddddf |||||} | ||| ||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)r5   expandreshape)r2   r;   batchnum_key_value_headsslenhead_dimr)   r)   r*   	repeat_kv3   s
    0rC           )modulequerykeyvalueattention_maskscalingdropoutkwargsc                 K   s   t || j}t || j}	t||dd| }
|d urf|d d d d d d d |jd f }|
| }
tjj|
dtj	d
|j}
tjj|
|| jd}
t|
|	}|dd }||
fS )Nr+   r   r,   )dimr-   )ptrainingr   )rC   num_key_value_groupsr!   matmul	transposer5   r   Z
functionalZsoftmaxr/   r.   r-   rK   rP   
contiguous)rE   rF   rG   rH   rI   rJ   rK   rL   
key_statesvalue_statesattn_weightscausal_maskattn_outputr)   r)   r*   eager_attention_forward?   s    
&rZ   c           
      C   s^   | j |j  }}||}||}| | t| |  }|| t||  }	|||	|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.
        position_ids (`torch.Tensor`, *optional*):
            Deprecated and unused.
        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.
    )r-   	unsqueezerotate_halfr.   )
qkcossinposition_idsZunsqueeze_dimZq_typeZk_typeZq_embedZk_embedr)   r)   r*   apply_rotary_pos_embY   s    

rb   c                 C   sH   | dd| j d d f }| d| j d d df }tj| |fddS )z*Rotates half the hidden dims of the input..Nr,   r+   rN   )r5   r!   cat)xx1Zx2r)   r)   r*   r\   u   s    r\   c                       s   e Zd ZdZdeee d fddZedddd	de	j
ee	j
e	j
f ee	j
 ee ee	j ee ee	j
ee	j
 f d
ddZ  ZS )Olmo2Attentionz=Multi-headed attention from 'Attention Is All You Need' paperNconfig	layer_idxc                    s   t    || _|| _t|d|j|j | _|j|j | _	| jd | _
|j| _d| _tj|j|j| j |jd| _tj|j|j| j |jd| _tj|j|j| j |jd| _tj|j| j |j|jd| _t|j| j |j| _t|j| j |j| _d S )NrB   g      TZbias)r   r   ri   rj   getattrr%   Znum_attention_headsrB   r@   rQ   rJ   attention_dropoutZ	is_causalr   LinearZattention_biasq_projk_projv_projo_projr   rms_norm_epsq_normk_normr$   ri   rj   r'   r)   r*   r      s,    
zOlmo2Attention.__init__past_key_valuepast_key_values4.58new_nameversion)r2   position_embeddingsrI   rx   cache_positionrL   r<   c                 K   s<  |j d d }g |d| jR }| | |}	| | |}
| |}|	|dd}	|
|dd}
||dd}|\}}t	|	|
||\}	}
|d ur|||d}|
|
|| j|\}
}t}| jjdkrt| jj }|| |	|
||f| jsdn| j| jd|\}}|jg |dR   }| |}||fS )Nr,   r   r+   )r`   r_   r~   eagerrD   )rK   rJ   )r5   rB   rt   ro   ru   rp   rq   viewrS   rb   updaterj   rZ   ri   Z_attn_implementationr   rP   rm   rJ   r>   rT   rr   )r$   r2   r}   rI   rx   r~   rL   Zinput_shapeZhidden_shapeZquery_statesrU   rV   r_   r`   Zcache_kwargsZattention_interfacerY   rW   r)   r)   r*   r3      s>    



zOlmo2Attention.forward)N)NN)r7   r8   r9   __doc__r   r   intr   r   r!   Tensorr4   r   
LongTensorr   r   r3   r:   r)   r)   r'   r*   rg   |   s     rg   c                       s$   e Zd Z fddZdd Z  ZS )Olmo2MLPc                    sr   t    || _|j| _|j| _tj| j| jdd| _tj| j| jdd| _tj| j| jdd| _	t
|j | _d S NFrk   )r   r   ri   r%   Zintermediate_sizer   rn   	gate_projup_proj	down_projr   Z
hidden_actact_fnr$   ri   r'   r)   r*   r      s    
zOlmo2MLP.__init__c                 C   s$   |  | | || | }|S )N)r   r   r   r   )r$   re   r   r)   r)   r*   r3      s     zOlmo2MLP.forward)r7   r8   r9   r   r3   r:   r)   r)   r'   r*   r      s   
r   c                       s   e Zd Zeed fddZedddddeje	ej e	ej
 e	e e	e e	ej
 e	eejejf  ee ejd
	ddZ  ZS )Olmo2DecoderLayerrh   c                    sR   t    |j| _t||d| _t|| _t|j|jd| _	t|j|jd| _
d S )Nrh   r&   )r   r   r%   rg   	self_attnr   mlpr   rs   post_attention_layernormpost_feedforward_layernormrv   r'   r)   r*   r      s    

zOlmo2DecoderLayer.__init__rw   rx   ry   rz   NF)	r2   rI   ra   rx   	use_cacher~   r}   rL   r<   c              
   K   s^   |}	| j f |||||||d|\}}
| |}|	| }|}	| |}| |}|	| }|S )N)r2   rI   ra   rx   r   r~   r}   )r   r   r   r   )r$   r2   rI   ra   rx   r   r~   r}   rL   Zresidual_r)   r)   r*   r3      s&    




zOlmo2DecoderLayer.forward)NNNFNN)r7   r8   r9   r   r   r   r   r!   r   r   r   r   boolr4   r   r   r3   r:   r)   r)   r'   r*   r      s&   	      r   c                       sD   e Zd ZU ejed< ded fddZe e	dd Z
  ZS )	Olmo2RotaryEmbeddinginv_freqNri   c                    s   t    t|dr:t|jtr:|jd|jd| _nd| _|j| _	|j| _
|| _t| j | _| | j|\}| _| jd|dd | j| _d S )Nrope_scaling	rope_typetypedefaultr   F)
persistent)r   r   hasattr
isinstancer   dictgetr   Zmax_position_embeddingsZmax_seq_len_cachedZoriginal_max_seq_lenri   r   Zrope_init_fnattention_scalingZregister_bufferr   Zoriginal_inv_freq)r$   ri   devicer   r'   r)   r*   r     s    
zOlmo2RotaryEmbedding.__init__c           
      C   s   | j d d d d f  |jd dd|j}|d d d d d f  }t|jjtrl|jjdkrl|jjnd}t	j
|dd^ | |  dd}t	j||fdd	}| | j }| | j }	||	fW  d    S 1 s0    Y  d S )
Nr   r,   r   ZmpscpuF)device_typeZenabledr+   rc   )r   floatr=   r5   r.   r   r   r   strr!   ZautocastrS   rd   r_   r   r`   )
r$   re   ra   Zinv_freq_expandedZposition_ids_expandedr   ZfreqsZembr_   r`   r)   r)   r*   r3     s    0&zOlmo2RotaryEmbedding.forward)N)r7   r8   r9   r!   r   __annotations__r   r   Zno_gradr   r3   r:   r)   r)   r'   r*   r     s
   

r   c                   @   sH   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dS )Olmo2PreTrainedModelri   modelTr   rx   )r2   
attentionsN)r7   r8   r9   r   r   Zbase_model_prefixZsupports_gradient_checkpointingZ_no_split_modulesZ_skip_keys_device_placementZ_supports_flash_attnZ_supports_sdpaZ_supports_flex_attnZ_can_compile_fullgraphZ_supports_attention_backendr   rg   Z_can_record_outputsr)   r)   r)   r*   r   (  s   
r   c                       st   e Zd Zed fddZeedeej	 eej
 eej	 ee eej eej	 ee ee ed	ddZ  ZS )	
Olmo2Modelr   c                    s   t     j| _ j| _t j j| j| _t	 fddt
 jD | _t j jd| _t d| _d| _|   d S )Nc                    s   g | ]}t  |qS r)   )r   ).0rj   r   r)   r*   
<listcomp>D      z'Olmo2Model.__init__.<locals>.<listcomp>r   r   F)r   r   Zpad_token_idZpadding_idx
vocab_sizer   Z	Embeddingr%   embed_tokensZ
ModuleListrangenum_hidden_layerslayersr   rs   normr   
rotary_embZgradient_checkpointing	post_initr   r'   r   r*   r   =  s    zOlmo2Model.__init__N)		input_idsrI   ra   rx   inputs_embedsr~   r   rL   r<   c              	   K   s   |d u |d uA rt d|d u r*| |}|rB|d u rBt| jd}|d u rz|d urZ| nd}	tj|	|	|jd  |jd}|d u r|	d}t
| j|||||d}
|}| ||}| jd | jj D ] }||f|
||||d|}q| |}t||dS )	Nz:You must specify exactly one of input_ids or inputs_embedsr   r   r   )r   )ri   Zinput_embedsrI   r~   rx   ra   )rI   ra   rx   r~   r}   )last_hidden_staterx   )
ValueErrorr   r	   ri   Zget_seq_lengthr!   Zaranger5   r   r[   r   r   r   r   r   r   )r$   r   rI   ra   rx   r   r~   r   rL   Zpast_seen_tokensrX   r2   r}   Zdecoder_layerr)   r)   r*   r3   M  sP    

	

zOlmo2Model.forward)NNNNNNN)r7   r8   r9   r   r   r   r   r   r!   r   r   r   FloatTensorr   r   r   r   r3   r:   r)   r)   r'   r*   r   ;  s*          r   c                       s   e Zd ZdgZddiZddgdgfiZ fddZeede	e
j e	e
j e	e
j e	e e	e
j e	e
j e	e e	e
j eee
jf ee ed
ddZ  ZS )Olmo2ForCausalLMzlm_head.weightlm_headZcolwise_repr2   logitsc                    s@   t  | t|| _|j| _tj|j|jdd| _| 	  d S r   )
r   r   r   r   r   r   rn   r%   r   r   r   r'   r)   r*   r     s
    
zOlmo2ForCausalLM.__init__Nr   )r   rI   ra   rx   r   labelsr   r~   logits_to_keeprL   r<   c
              
   K   s   | j f |||||||d|
}|j}t|	tr<t|	 dn|	}| |dd|ddf }d}|dur| jf ||| jjd|
}t	|||j
|j|jdS )a  
        Example:

        ```python
        >>> from transformers import AutoTokenizer, Olmo2ForCausalLM

        >>> model = Olmo2ForCausalLM.from_pretrained("meta-olmo2/Olmo2-2-7b-hf")
        >>> tokenizer = AutoTokenizer.from_pretrained("meta-olmo2/Olmo2-2-7b-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   rI   ra   rx   r   r   r~   N)r   r   r   )lossr   rx   r2   r   )r   r   r   r   slicer   Zloss_functionri   r   r   rx   r2   r   )r$   r   rI   ra   rx   r   r   r   r~   r   rL   outputsr2   Zslice_indicesr   r   r)   r)   r*   r3     s0     zOlmo2ForCausalLM.forward)	NNNNNNNNr   )r7   r8   r9   Z_tied_weights_keysZ_tp_planZ_pp_planr   r   r   r   r!   r   r   r   r   r   r   r   r   r   r   r3   r:   r)   r)   r'   r*   r     s8   	         r   )r   r   r   )rD   )Nr   );typingr   r   r   r!   Ztorch.nnr   Ztransformers.utils.genericr   Zactivationsr   Zcache_utilsr   r	   Z
generationr
   Zintegrationsr   Zmasking_utilsr   Zmodeling_layersr   Zmodeling_outputsr   r   Zmodeling_rope_utilsr   r   Zmodeling_utilsr   r   Zprocessing_utilsr   utilsr   r   Zutils.deprecationr   Zutils.genericr   Zconfiguration_olmo2r   Moduler   r   r   rC   r   rZ   rb   r\   rg   r   r   r   r   r   r   __all__r)   r)   r)   r*   <module>   sX    
M,#NK