o
    wi                     @   sr  d dl mZmZmZ d dl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 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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% ddl&m'Z'm(Z( ddl)m*Z* e+e,Z-G dd dej.Z/edG dd dej.Z0G dd dej.Z1dd Z2dCddZ3dej4d e5d!ej4fd"d#Z6	$dDd%ej.d&ej4d'ej4d(ej4d)eej4 d*e7d+e7fd,d-Z8G d.d/ d/ej.Z9G d0d1 d1eZ:eG d2d3 d3e#Z;eG d4d5 d5e;Z<G d6d7 d7ee'Z=ed8d9G d:d; d;e;eZ>ed8d9G d<d= d=e;Z?ed8d9G d>d? d?e;Z@ed8d9G d@dA dAe;ZAg dBZBdS )E    )CallableOptionalUnionN)nn)auto_docstringlogging   )ACT2FN)CacheDynamicCache)GenerationMixin)use_kernel_forward_from_hub)create_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPastQuestionAnsweringModelOutput SequenceClassifierOutputWithPastTokenClassifierOutput)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)
LossKwargscan_return_tuple   )ArceeConfigc                       s$   e Zd Z fddZdd Z  ZS )ArceeMLPc                    s`   t    || _|j| _|j| _tj| j| j|jd| _tj| j| j|jd| _	t
|j | _d S )Nbias)super__init__confighidden_sizeintermediate_sizer   Linearmlp_biasup_proj	down_projr	   
hidden_actact_fnselfr$   	__class__ e/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/transformers/models/arcee/modeling_arcee.pyr#   6   s   
zArceeMLP.__init__c                 C   s   |  | | |S N)r*   r,   r)   )r.   xr1   r1   r2   forward?   s   zArceeMLP.forward)__name__
__module____qualname__r#   r5   __classcell__r1   r1   r/   r2   r   5   s    	r   RMSNormc                       s.   e Zd Zd fdd	Zdd Zdd Z  ZS )	ArceeRMSNormư>c                    s&   t    tt|| _|| _dS )z;
        ArceeRMSNorm is equivalent to T5LayerNorm
        N)r"   r#   r   	Parametertorchonesweightvariance_epsilon)r.   r%   epsr/   r1   r2   r#   E   s   

zArceeRMSNorm.__init__c                 C   sJ   |j }|tj}|djddd}|t|| j  }| j|| S )N   T)keepdim)	dtypetor>   float32powmeanrsqrtrA   r@   )r.   hidden_statesinput_dtypevariancer1   r1   r2   r5   M   s
   zArceeRMSNorm.forwardc                 C   s   t | jj d| j S )Nz, eps=)tupler@   shaperA   r.   r1   r1   r2   
extra_reprT   s   zArceeRMSNorm.extra_repr)r<   )r6   r7   r8   r#   r5   rR   r9   r1   r1   r/   r2   r;   C   s    r;   c                       s8   e Zd Zddef fddZe edd Z  Z	S )ArceeRotaryEmbeddingNr$   c                    s   t    t|dr|jd u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defaultinv_freqF)
persistent)r"   r#   hasattrrT   getrU   max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenr$   r   rope_init_fnattention_scalingregister_bufferrX   original_inv_freq)r.   r$   devicerX   r/   r1   r2   r#   Y   s   
zArceeRotaryEmbedding.__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r6|jjdkr6|jjnd}t	j
|dd+ | |  dd}t	j||fdd	}| | j }| | j }	W d    n1 smw   Y  |j|jd
|	j|jd
fS )Nr   rD   r   mpscpuF)device_typeenabledrC   dim)rF   )rX   floatexpandrP   rG   rc   
isinstancerV   strr>   autocast	transposecatcosr`   sinrF   )
r.   r4   position_idsinv_freq_expandedposition_ids_expandedrf   freqsembrq   rr   r1   r1   r2   r5   j   s   0&zArceeRotaryEmbedding.forwardr3   )
r6   r7   r8   r   r#   r>   no_gradr   r5   r9   r1   r1   r/   r2   rS   X   s
    rS   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..NrD   rC   rh   )rP   r>   rp   )r4   x1x2r1   r1   r2   rotate_halfz   s   r{   c                 C   sD   | |}| |}| | 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.
    )	unsqueezer{   )qkrq   rr   rs   unsqueeze_dimq_embedk_embedr1   r1   r2   apply_rotary_pos_emb   s
   

r   rL   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)rP   rk   reshape)rL   r   batchnum_key_value_headsslenhead_dimr1   r1   r2   	repeat_kv   s
   0r           modulequerykeyvalueattention_maskscalingdropoutc                 K   s   t || j}t || j}	t||dd| }
|d ur3|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 )NrC   r   rD   )ri   rF   )ptrainingr   )r   num_key_value_groupsr>   matmulro   rP   r   
functionalsoftmaxrH   rG   rF   r   r   
contiguous)r   r   r   r   r   r   r   kwargs
key_statesvalue_statesattn_weightscausal_maskattn_outputr1   r1   r2   eager_attention_forward   s   
&r   c                       s   e Zd ZdZdedef fddZ		ddejde	ejejf d	e
ej d
e
e de
ej dee de	eje
ej e
e	ej  f fddZ  ZS )ArceeAttentionz=Multi-headed attention from 'Attention Is All You Need' paperr$   	layer_idxc                    s   t    || _|| _t|d|j|j | _|j|j | _	| jd | _
|j| _d| _tj|j|j| j |jd| _tj|j|j| j |jd| _tj|j|j| j |jd| _tj|j| j |j|jd| _d S )Nr   g      Tr    )r"   r#   r$   r   getattrr%   num_attention_headsr   r   r   r   attention_dropout	is_causalr   r'   attention_biasq_projk_projv_projo_projr.   r$   r   r/   r1   r2   r#      s(   
zArceeAttention.__init__NrL   position_embeddingsr   past_key_valuecache_positionr   r   c                 K   s$  |j d d }g |d| jR }| ||dd}	| ||dd}
| ||dd}|\}}t|	|
||\}	}
|d urW|||d}||
|| j	|\}
}t
}| jjdkret| jj }|| |	|
||f| jsqdn| j| jd|\}}|jg |dR   }| |}||fS )NrD   r   rC   )rr   rq   r   eagerr   )r   r   )rP   r   r   viewro   r   r   r   updater   r   r$   _attn_implementationr   r   r   r   r   r   r   )r.   rL   r   r   r   r   r   input_shapehidden_shapequery_statesr   r   rq   rr   cache_kwargsattention_interfacer   r   r1   r1   r2   r5      s8   	

zArceeAttention.forward)NN)r6   r7   r8   __doc__r   intr#   r>   TensorrO   r   r
   
LongTensorr   r   r5   r9   r1   r1   r/   r2   r      s(    r   c                       s   e Zd Zdedef fddZ							ddejdeej d	eej	 d
ee
 dee dee deej	 deeejejf  dee deejeeejejf  f fddZ  ZS )ArceeDecoderLayerr$   r   c                    sR   t    |j| _t||d| _t|| _t|j|jd| _	t|j|jd| _
d S )N)r$   r   rB   )r"   r#   r%   r   	self_attnr   mlpr;   rms_norm_epsinput_layernormpost_attention_layernormr   r/   r1   r2   r#   	  s   

zArceeDecoderLayer.__init__NFrL   r   rs   r   output_attentions	use_cacher   r   r   r   c	                 K   st   |}
|  |}| jd||||||||d|	\}}|
| }|}
| |}| |}|
| }|f}|r8||f7 }|S )N)rL   r   rs   r   r   r   r   r   r1   )r   r   r   r   )r.   rL   r   rs   r   r   r   r   r   r   residualself_attn_weightsoutputsr1   r1   r2   r5     s.   
	



zArceeDecoderLayer.forward)NNNFFNN)r6   r7   r8   r   r   r#   r>   r   r   r   r
   boolrO   r   r   FloatTensorr5   r9   r1   r1   r/   r2   r     s<    	
r   c                   @   sL   e Zd ZeZdZdZdgZdgZdZ	dZ
dZdZdZdZdZdZdd ZdS )ArceePreTrainedModelmodelTr   past_key_valuesc                 C   s   | j j}t|tjr"|jjjd|d |jd ur |jj	  d S d S t|tj
rC|jjjd|d |jd urA|jj|j 	  d S d S t|trQ|jjd d S d S )Nr   )rJ   stdg      ?)r$   initializer_rangerl   r   r'   r@   datanormal_r!   zero_	Embeddingpadding_idxr;   fill_)r.   r   r   r1   r1   r2   _init_weightsM  s   


z"ArceePreTrainedModel._init_weightsN)r6   r7   r8   r   config_classbase_model_prefixsupports_gradient_checkpointing_no_split_modules_skip_keys_device_placement_supports_flash_attn_3_supports_flash_attn_2_supports_sdpa_supports_flex_attn_supports_cache_class_supports_quantized_cache_supports_static_cache_supports_attention_backendr   r1   r1   r1   r2   r   =  s    r   c                       s   e Zd Zdef fddZdd Zdd Zee									dd	e	e
j d
e	e
j de	e
j de	e de	e
j de	e de	e de	e de	e
j dee defddZ  ZS )
ArceeModelr$   c                    s   t     j| _ j| _t j j| j| _t	 fddt
 jD | _t j jd| _t d| _d| _|   d S )Nc                    s   g | ]}t  |qS r1   )r   ).0r   r$   r1   r2   
<listcomp>d  s    z'ArceeModel.__init__.<locals>.<listcomp>r   r   F)r"   r#   pad_token_idr   
vocab_sizer   r   r%   embed_tokens
ModuleListrangenum_hidden_layerslayersr;   r   normrS   
rotary_embgradient_checkpointing	post_initr-   r/   r   r2   r#   ]  s   zArceeModel.__init__c                 C      | j S r3   r   rQ   r1   r1   r2   get_input_embeddingsm     zArceeModel.get_input_embeddingsc                 C   
   || _ d S r3   r   r.   r   r1   r1   r2   set_input_embeddingsp     
zArceeModel.set_input_embeddingsN	input_idsr   rs   r   inputs_embedsr   r   output_hidden_statesr   flash_attn_kwargsr   c
                 K   s  |d ur|n| j j}|d ur|n| j j}|d ur|n| j j}|d u |d uA r*td| jr9| jr9|r9td d}t	|t
d tfsFtd|d u rO| |}|rX|d u rXt }|	d u rt|d urd| nd}tj|||jd  |jd}	|d u r}|	d}t| j |||	||d}|}| ||}|rd	nd }|rd	nd }| jd | j j D ]&}|r||f7 }||f||||||	|d
|
}|d }|r||d f7 }q| |}|r||f7 }t||r|nd ||dS )Nz:You must specify exactly one of input_ids or inputs_embedszX`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.FzBThe `past_key_values` should be either a `Cache` object or `None`.r   r   rc   )r$   input_embedsr   r   r   rs   r1   )r   rs   r   r   r   r   r   )last_hidden_stater   rL   
attentions)r$   r   r   r   
ValueErrorr   r   loggerwarning_oncerl   rV   r
   r   r   get_seq_lengthr>   arangerP   rc   r|   r   r   r   r   r   r   )r.   r   r   rs   r   r   r   r   r   r   r   past_seen_tokensr   rL   r   all_hidden_statesall_self_attnsdecoder_layerlayer_outputsr1   r1   r2   r5   s  s   

	
	


zArceeModel.forward	NNNNNNNNN)r6   r7   r8   r   r#   r   r   r   r   r   r>   r   r   r
   r   r   r   r   r   r5   r9   r1   r1   r/   r2   r   [  sL    	
r   c                   @   s   e Zd ZdS )KwargsForCausalLMN)r6   r7   r8   r1   r1   r1   r2   r    s    r  zarcee-ai/AFM-4.5B)
checkpointc                       s
  e Zd ZdgZddiZddgdgfiZ fddZdd	 Zd
d Zdd Z	dd Z
dd Zdd Zee											d%deej deej deej dee deej deej dee dee dee deej d eeejf d!ee d"efd#d$Z  ZS )&ArceeForCausalLMzlm_head.weightlm_headcolwise_reprL   logitsc                    s@   t  | t|| _|j| _tj|j|jdd| _| 	  d S NFr    )
r"   r#   r   r   r   r   r'   r%   r  r   r-   r/   r1   r2   r#     s
   
zArceeForCausalLM.__init__c                 C      | j jS r3   r   r   rQ   r1   r1   r2   r        z%ArceeForCausalLM.get_input_embeddingsc                 C      || j _d S r3   r  r   r1   r1   r2   r        z%ArceeForCausalLM.set_input_embeddingsc                 C   r   r3   r  rQ   r1   r1   r2   get_output_embeddings  r   z&ArceeForCausalLM.get_output_embeddingsc                 C   r   r3   r  )r.   new_embeddingsr1   r1   r2   set_output_embeddings  r   z&ArceeForCausalLM.set_output_embeddingsc                 C   r   r3   r   )r.   decoderr1   r1   r2   set_decoder  r   zArceeForCausalLM.set_decoderc                 C   r   r3   r  rQ   r1   r1   r2   get_decoder  r   zArceeForCausalLM.get_decoderNr   r   r   rs   r   r   labelsr   r   r   r   logits_to_keepr   r   c                 K   s   |dur|n| j j}|	dur|	n| j j}	| jd||||||||	|
d	|}|j}t|tr4t| dn|}| |dd|ddf }d}|durX| j	d||| j j
d|}t|||j|j|jdS )at  
        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, ArceeForCausalLM

        >>> model = ArceeForCausalLM.from_pretrained("meta-arcee/Arcee-2-7b-hf")
        >>> tokenizer = AutoTokenizer.from_pretrained("meta-arcee/Arcee-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."
        ```N)	r   r   rs   r   r   r   r   r   r   )r  r#  r   lossr  r   rL   r  r1   )r$   r   r   r   r  rl   r   slicer  loss_functionr   r   r   rL   r  )r.   r   r   rs   r   r   r#  r   r   r   r   r$  r   r   rL   slice_indicesr  r&  r1   r1   r2   r5     s:   '
zArceeForCausalLM.forward)NNNNNNNNNNr   )r6   r7   r8   _tied_weights_keys_tp_plan_pp_planr#   r   r   r  r  r!  r"  r   r   r   r>   r   r   r
   r   r   r   r   r   r  r   r5   r9   r1   r1   r/   r2   r    sf    		
r  c                          e Zd Z fddZdd Zdd Zee									ddee	j
 d	ee	j d
ee	j
 dee dee	j dee	j
 dee dee dee defddZ  ZS )ArceeForSequenceClassificationc                    s@   t  | |j| _t|| _tj|j| jdd| _| 	  d S r  )
r"   r#   
num_labelsr   r   r   r'   r%   scorer   r-   r/   r1   r2   r#   K  s
   
z'ArceeForSequenceClassification.__init__c                 C   r  r3   r  rQ   r1   r1   r2   r   T  r  z3ArceeForSequenceClassification.get_input_embeddingsc                 C   r  r3   r  r   r1   r1   r2   r   W  r  z3ArceeForSequenceClassification.set_input_embeddingsNr   r   rs   r   r   r#  r   r   r   r   c
              
   C   s(  | j ||||||||	d}
|
j}| |}|dur|jd }n|jd }| jjdu r2|dkr2td| jjdu r;d}n1|dur`|| jjk|jt	j
}t	j|jd |jt	j
d}|| d}nd}t| jj d |t	j||jd	|f }d}|dur| j|||| jd
}t|||
j|
j|
jdS )  
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        r   rs   r   r   r   r   r   Nr   r   z=Cannot handle batch sizes > 1 if no padding token is defined.rD   )rc   rF   z will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`r   )r  r#  pooled_logitsr$   r%  )r   r  r0  rP   r$   r   r  rG   rc   r>   int32r  argmaxr  r  r0   r6   r(  r   r   rL   r  )r.   r   r   rs   r   r   r#  r   r   r   transformer_outputsrL   r  
batch_sizelast_non_pad_tokennon_pad_masktoken_indicesr3  r&  r1   r1   r2   r5   Z  sL   


z&ArceeForSequenceClassification.forwardr  )r6   r7   r8   r#   r   r   r   r   r   r>   r   r   r
   r   r   r   r5   r9   r1   r1   r/   r2   r.  I  sH    		
r.  c                       s   e Zd ZdZ fddZdd Zdd Zee									dd	e	e
j d
e	e
j de	e
j de	e de	e
j de	e
j de	e
j de	e de	e defddZ  ZS )ArceeForQuestionAnsweringtransformerc                    s2   t  | t|| _t|jd| _|   d S )NrC   )	r"   r#   r   r<  r   r'   r%   
qa_outputsr   r-   r/   r1   r2   r#     s   
z"ArceeForQuestionAnswering.__init__c                 C   r  r3   r<  r   rQ   r1   r1   r2   r     r  z.ArceeForQuestionAnswering.get_input_embeddingsc                 C   r  r3   r>  r   r1   r1   r2   r     r  z.ArceeForQuestionAnswering.set_input_embeddingsNr   r   rs   r   r   start_positionsend_positionsr   r   r   c
              	   K   s   | j |||||||	d}|j}| |}|jddd\}}|d }|d }d }|d urA|d urA| j||||fi |
}t||||j|j	dS )N)r   rs   r   r   r   r   r   rD   rh   )r&  start_logits
end_logitsrL   r  )
r<  r  r=  splitsqueezer   r(  r   rL   r  )r.   r   r   rs   r   r   r?  r@  r   r   r   r   sequence_outputr  rA  rB  r&  r1   r1   r2   r5     s0   

z!ArceeForQuestionAnswering.forwardr  )r6   r7   r8   r   r#   r   r   r   r   r   r>   r   r   r
   r   r   r   r5   r9   r1   r1   r/   r2   r;    sJ    	
r;  c                       r-  )ArceeForTokenClassificationc                    s|   t  | |j| _t|| _t|dd d ur|j}nt|dd d ur'|j}nd}t	|| _
t|j|j| _|   d S )Nclassifier_dropouthidden_dropoutg?)r"   r#   r/  r   r   r   rG  rH  r   Dropoutr   r'   r%   r0  r   )r.   r$   rG  r/   r1   r2   r#     s   
z$ArceeForTokenClassification.__init__c                 C   r  r3   r  rQ   r1   r1   r2   r     r  z0ArceeForTokenClassification.get_input_embeddingsc                 C   r  r3   r  r   r1   r1   r2   r     r  z0ArceeForTokenClassification.set_input_embeddingsNr   r   rs   r   r   r#  r   r   r   r   c
              
   C   sd   | j ||||||||	d}
|
j}| |}| |}d}|dur(| ||| j}t|||
j|
jdS )r1  r2  N)r&  r  rL   r  )	r   r  r   r0  r(  r$   r   rL   r  )r.   r   r   rs   r   r   r#  r   r   r   r   rE  r  r&  r1   r1   r2   r5     s,   


z#ArceeForTokenClassification.forwardr  )r6   r7   r8   r#   r   r   r   r   r   r>   r   r   r
   r   r   r   r5   r9   r1   r1   r/   r2   rF    sH    	
rF  )r  r;  r.  rF  r   r   )Nr   )r   )Ctypingr   r   r   r>   r   transformers.utilsr   r   activationsr	   cache_utilsr
   r   
generationr   integrationsr   masking_utilsr   modeling_flash_attention_utilsr   modeling_layersr   modeling_outputsr   r   r   r   r   modeling_rope_utilsr   r   modeling_utilsr   r   processing_utilsr   utilsr   r   configuration_arceer   
get_loggerr6   r  Moduler   r;   rS   r{   r   r   r   r   rj   r   r   r   r   r   r  r  r.  r;  rF  __all__r1   r1   r1   r2   <module>   sr   
"

F5}lV>F