o
    wi                     @   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	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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$m%Z%m&Z&m'Z'm(Z( ddl)m*Z* e(+e,Z-edG dd dej.Z/G dd dej.Z0dd Z1dBddZ2dej3de4dej3fdd Z5	!dCd"ej.d#ej3d$ej3d%ej3d&eej3 d'e6d(e6fd)d*Z7G d+d, d,ej.Z8G d-d. d.eZ9e&G d/d0 d0e!Z:G d1d2 d2ej.Z;e&G d3d4 d4e:Z<G d5d6 d6ee%Z=e&G d7d8 d8e:eZ>e&d9d:G d;d< d<e:Z?e&G d=d> d>e:Z@e&G d?d@ d@e:ZAg dAZBdS )D    )CallableOptionalUnionN)nn   )ACT2FN)CacheDynamicCache)GenerationMixin)use_kernel_forward_from_hub)create_causal_mask!create_sliding_window_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPastQuestionAnsweringModelOutput SequenceClassifierOutputWithPastTokenClassifierOutput)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)
LossKwargsauto_docstringcan_return_tuplelogging   )Qwen3ConfigRMSNormc                       s.   e Zd Zd fdd	Zdd Zdd Z  ZS )	Qwen3RMSNormư>c                    s&   t    tt|| _|| _dS )z;
        Qwen3RMSNorm is equivalent to T5LayerNorm
        N)super__init__r   	Parametertorchonesweightvariance_epsilon)selfhidden_sizeeps	__class__ e/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/transformers/models/qwen3/modeling_qwen3.pyr$   5   s   

zQwen3RMSNorm.__init__c                 C   sJ   |j }|tj}|djddd}|t|| j  }| j|| S )N   T)keepdim)	dtypetor&   float32powmeanrsqrtr)   r(   )r*   hidden_statesinput_dtypevariancer/   r/   r0   forward=   s
   zQwen3RMSNorm.forwardc                 C   s   t | jj d| j S )Nz, eps=)tupler(   shaper)   r*   r/   r/   r0   
extra_reprD   s   zQwen3RMSNorm.extra_repr)r"   )__name__
__module____qualname__r$   r=   rA   __classcell__r/   r/   r-   r0   r!   3   s    r!   c                       s$   e Zd Z fddZdd Z  ZS )Qwen3MLPc                    sr   t    || _|j| _|j| _tj| j| jdd| _tj| j| jdd| _tj| j| jdd| _	t
|j | _d S NFbias)r#   r$   configr+   intermediate_sizer   Linear	gate_projup_proj	down_projr   
hidden_actact_fnr*   rJ   r-   r/   r0   r$   I   s   
zQwen3MLP.__init__c                 C   s$   |  | | || | }|S N)rO   rQ   rM   rN   )r*   xrO   r/   r/   r0   r=   S   s    zQwen3MLP.forward)rB   rC   rD   r$   r=   rE   r/   r/   r-   r0   rF   H   s    
rF   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..Nr2   r1   dim)r?   r&   cat)rT   x1x2r/   r/   r0   rotate_halfX   s   rZ   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.
    )	unsqueezerZ   )qkcossinposition_idsunsqueeze_dimq_embedk_embedr/   r/   r0   apply_rotary_pos_emb_   s
   

rd   r:   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)r?   expandreshape)r:   re   batchnum_key_value_headsslenhead_dimr/   r/   r0   	repeat_kvz   s
   0rm           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 )Nr1   r   r2   )rV   r4   )ptrainingr   )rm   num_key_value_groupsr&   matmul	transposer?   r   
functionalsoftmaxr6   r5   r4   ru   rx   
contiguous)ro   rp   rq   rr   rs   rt   ru   kwargs
key_statesvalue_statesattn_weightscausal_maskattn_outputr/   r/   r0   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 )Qwen3Attentionz=Multi-headed attention from 'Attention Is All You Need' paperrJ   	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d| _t| j|jd| _|j| dkr|j| _d S d | _d S )Nrl   g      TrH   r,   sliding_attention)r#   r$   rJ   r   getattrr+   num_attention_headsrl   rj   ry   rt   attention_dropout	is_causalr   rL   attention_biasq_projk_projv_projo_projr!   rms_norm_epsq_normk_normlayer_typessliding_windowr*   rJ   r   r-   r/   r0   r$      s.   
$zQwen3Attention.__init__Nr:   position_embeddingsrs   past_key_valuecache_positionr   rf   c                 K   s4  |j d d }g |d| jR }| | ||dd}	| | ||dd}
| ||dd}|\}}t	|	|
||\}	}
|d ur]|||d}|
|
|| j|\}
}t}| jjdkrkt| jj }|| |	|
||f| jswdn| j| j| jd|\}}|jg |dR   }| |}||fS )Nr2   r   r1   )r_   r^   r   eagerrn   )ru   rt   r   )r?   rl   r   r   viewr{   r   r   r   rd   updater   r   rJ   _attn_implementationr   rx   r   rt   r   rh   r~   r   )r*   r:   r   rs   r   r   r   input_shapehidden_shapequery_statesr   r   r^   r_   cache_kwargsattention_interfacer   r   r/   r/   r0   r=      s:   		

zQwen3Attention.forward)NN)rB   rC   rD   __doc__r   intr$   r&   Tensorr>   r   r   
LongTensorr   r   r=   rE   r/   r/   r-   r0   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 )Qwen3DecoderLayerrJ   r   c                    s^   t    |j| _t||d| _t|| _t|j|jd| _	t|j|jd| _
|j| | _d S )N)rJ   r   r   )r#   r$   r+   r   	self_attnrF   mlpr!   r   input_layernormpost_attention_layernormr   attention_typer   r-   r/   r0   r$      s   

zQwen3DecoderLayer.__init__NFr:   rs   r`   r   output_attentions	use_cacher   r   r   rf   c	                 K   st   |}
|  |}| jd||||||||d|	\}}|
| }|}
| |}| |}|
| }|f}|r8||f7 }|S )N)r:   rs   r`   r   r   r   r   r   r/   )r   r   r   r   )r*   r:   rs   r`   r   r   r   r   r   r   residualself_attn_weightsoutputsr/   r/   r0   r=      s.   
	



zQwen3DecoderLayer.forward)NNNFFNN)rB   rC   rD   r   r   r$   r&   r   r   r   r   boolr>   r   r   FloatTensorr=   rE   r/   r/   r-   r0   r      s<    	
r   c                   @   sL   e Zd ZeZdZdZdgZdgZdZ	dZ
dZdZdZdZdZdZdd ZdS )Qwen3PreTrainedModelmodelTr   past_key_valuesc                 C   s   | j j}t|tjr"|jjjd|d |jd ur |jj	  d S d S t|tj
rC|jjjd|d |jd urA|jj|j 	  d S d S t|trQ|jjd d S d S )Nrn   )r8   stdg      ?)rJ   initializer_range
isinstancer   rL   r(   datanormal_rI   zero_	Embeddingpadding_idxr!   fill_)r*   ro   r   r/   r/   r0   _init_weights0  s   


z"Qwen3PreTrainedModel._init_weightsN)rB   rC   rD   r   config_classbase_model_prefixsupports_gradient_checkpointing_no_split_modules_skip_keys_device_placement_supports_flash_attn_3_supports_flash_attn_2_supports_sdpa_supports_flex_attn_supports_cache_class_supports_quantized_cache_supports_static_cache_supports_attention_backendr   r/   r/   r/   r0   r      s    r   c                       s8   e Zd Zddef fddZe edd Z  Z	S )Qwen3RotaryEmbeddingNrJ   c                    s   t    t|dr|jd ur|jd|jd| _nd| _|j| _|j| _|| _	t
| j | _| | j	|\}| _| jd|dd | j| _d S )Nrope_scaling	rope_typetypedefaultinv_freqF)
persistent)r#   r$   hasattrr   getr   max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrJ   r   rope_init_fnattention_scalingregister_bufferr   original_inv_freq)r*   rJ   devicer   r-   r/   r0   r$   ?  s   
zQwen3RotaryEmbedding.__init__c           
      C   s   | j d d d d f  |jd dd|j}|d d d d d f  }t|jjtr6|jjdkr6|jjnd}t	j
|dd+ | |  dd}t	j||fdd	}| | j }| | j }	W d    n1 smw   Y  |j|jd
|	j|jd
fS )Nr   r2   r   mpscpuF)device_typeenabledr1   rU   )r4   )r   floatrg   r?   r5   r   r   r   strr&   autocastr{   rW   r^   r   r_   r4   )
r*   rT   r`   inv_freq_expandedposition_ids_expandedr   freqsembr^   r_   r/   r/   r0   r=   P  s   0&zQwen3RotaryEmbedding.forwardrS   )
rB   rC   rD   r   r$   r&   no_gradr   r=   rE   r/   r/   r-   r0   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 )
Qwen3ModelrJ   c                    s   t     j| _ j| _t j j| j| _t	 fddt
 jD | _t j jd| _t d| _d| _d| jjv | _|   d S )Nc                    s   g | ]}t  |qS r/   )r   ).0r   rJ   r/   r0   
<listcomp>i  s    z'Qwen3Model.__init__.<locals>.<listcomp>r   r   Fr   )r#   r$   pad_token_idr   
vocab_sizer   r   r+   embed_tokens
ModuleListrangenum_hidden_layerslayersr!   r   normr   
rotary_embgradient_checkpointingrJ   r   has_sliding_layers	post_initrR   r-   r   r0   r$   b  s   zQwen3Model.__init__c                 C      | j S rS   r   r@   r/   r/   r0   get_input_embeddingss     zQwen3Model.get_input_embeddingsc                 C   
   || _ d S rS   r   r*   rr   r/   r/   r0   set_input_embeddingsv     
zQwen3Model.set_input_embeddingsN	input_idsrs   r`   r   inputs_embedsr   r   output_hidden_statesr   flash_attn_kwargsrf   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	| }ts| j |||	||d}d	tdi |i}| jrtdi ||d
< |}| ||}|rdnd }|rdnd }| jd | j j D ])}|r||f7 }||f||j |||||	|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   r   )rJ   input_embedsrs   r   r   r`   full_attentionr   r/   )rs   r`   r   r   r   r   r   )last_hidden_stater   r:   
attentions)rJ   r   r  r   
ValueErrorr   rx   loggerwarning_oncer   r   r   r   r	   get_seq_lengthr&   aranger?   r   r[   dictr   r   r   r   r   r   r   r   r   )r*   r  rs   r`   r   r  r   r   r  r   r  past_seen_tokenscausal_mask_mappingmask_kwargsr:   r   all_hidden_statesall_self_attnsdecoder_layerlayer_outputsr/   r/   r0   r=   y  s   



	


zQwen3Model.forward	NNNNNNNNN)rB   rC   rD   r   r$   r   r  r   r   r   r&   r   r   r   r   r   r   r   r   r=   rE   r/   r/   r-   r0   r   `  sL    	
r   c                   @   s   e Zd ZdS )KwargsForCausalLMN)rB   rC   rD   r/   r/   r/   r0   r    s    r  c                       s
  e Zd ZdgZddiZddgdgfiZ fddZdd	 Zd
d Zdd Z	dd Z
dd Zdd Zee											d%deej deej deej dee deej deej dee dee dee deej d eeejf d!ee d"efd#d$Z  ZS )&Qwen3ForCausalLMzlm_head.weightlm_headcolwise_repr:   logitsc                    s@   t  | t|| _|j| _tj|j|jdd| _| 	  d S rG   )
r#   r$   r   r   r   r   rL   r+   r  r   rR   r-   r/   r0   r$     s
   
zQwen3ForCausalLM.__init__c                 C      | j jS rS   r   r   r@   r/   r/   r0   r        z%Qwen3ForCausalLM.get_input_embeddingsc                 C      || j _d S rS   r"  r  r/   r/   r0   r       z%Qwen3ForCausalLM.set_input_embeddingsc                 C   r   rS   r  r@   r/   r/   r0   get_output_embeddings  r   z&Qwen3ForCausalLM.get_output_embeddingsc                 C   r  rS   r&  )r*   new_embeddingsr/   r/   r0   set_output_embeddings  r  z&Qwen3ForCausalLM.set_output_embeddingsc                 C   r  rS   r   )r*   decoderr/   r/   r0   set_decoder  r  zQwen3ForCausalLM.set_decoderc                 C   r   rS   r*  r@   r/   r/   r0   get_decoder
  r   zQwen3ForCausalLM.get_decoderNr   r  rs   r`   r   r  labelsr   r   r  r   logits_to_keepr   rf   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 )a^  
        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, Qwen3ForCausalLM

        >>> model = Qwen3ForCausalLM.from_pretrained("Qwen/Qwen3-8B")
        >>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")

        >>> 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  rs   r`   r   r  r   r   r  r   )r   r.  r   lossr   r   r:   r  r/   )rJ   r   r  r   r  r   r   slicer  loss_functionr   r   r   r:   r  )r*   r  rs   r`   r   r  r.  r   r   r  r   r/  r   r   r:   slice_indicesr   r1  r/   r/   r0   r=     s:   '
zQwen3ForCausalLM.forward)NNNNNNNNNNr   )rB   rC   rD   _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   r=   rE   r/   r/   r-   r0   r    sf    		
r  a  
    The Qwen3 Model transformer with a sequence classification head on top (linear layer).

    [`Qwen3ForSequenceClassification`] uses the last token in order to do the classification, as other causal models
    (e.g. GPT-2) do.

    Since it does classification on the last token, it requires to know the position of the last token. If a
    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
    each row of the batch).
    )custom_introc                          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 )Qwen3ForSequenceClassificationc                    s@   t  | |j| _t|| _tj|j| jdd| _| 	  d S rG   )
r#   r$   
num_labelsr   r   r   rL   r+   scorer   rR   r-   r/   r0   r$   h  s
   
z'Qwen3ForSequenceClassification.__init__c                 C   r!  rS   r"  r@   r/   r/   r0   r   q  r#  z3Qwen3ForSequenceClassification.get_input_embeddingsc                 C   r$  rS   r"  r  r/   r/   r0   r  t  r%  z3Qwen3ForSequenceClassification.set_input_embeddingsNr  rs   r`   r   r  r.  r   r   r  rf   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).
        rs   r`   r   r  r   r   r  Nr   r   z=Cannot handle batch sizes > 1 if no padding token is defined.r2   )r   r4   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_logitsrJ   r0  )r   r  r<  r?   rJ   r   r  r5   r   r&   int32r  argmaxr  r  r.   rB   r3  r   r   r:   r  )r*   r  rs   r`   r   r  r.  r   r   r  transformer_outputsr:   r   
batch_sizelast_non_pad_tokennon_pad_masktoken_indicesr?  r1  r/   r/   r0   r=   w  sL   


z&Qwen3ForSequenceClassification.forwardr  )rB   rC   rD   r$   r   r  r   r   r   r&   r   r   r   r   r   r   r=   rE   r/   r/   r-   r0   r:  Y  sH    		
r:  c                       r9  )Qwen3ForTokenClassificationc                    s|   t  | |j| _t|| _t|dd d ur|j}nt|dd d ur'|j}nd}t	|| _
t|j|j| _|   d S )Nclassifier_dropouthidden_dropoutg?)r#   r$   r;  r   r   r   rH  rI  r   Dropoutru   rL   r+   r<  r   )r*   rJ   rH  r-   r/   r0   r$     s   
z$Qwen3ForTokenClassification.__init__c                 C   r!  rS   r"  r@   r/   r/   r0   r     r#  z0Qwen3ForTokenClassification.get_input_embeddingsc                 C   r$  rS   r"  r  r/   r/   r0   r    r%  z0Qwen3ForTokenClassification.set_input_embeddingsNr  rs   r`   r   r  r.  r   r   r  rf   c
              
   C   sd   | j ||||||||	d}
|
j}| |}| |}d}|dur(| ||| j}t|||
j|
jdS )r=  r>  N)r1  r   r:   r  )	r   r  ru   r<  r3  rJ   r   r:   r  )r*   r  rs   r`   r   r  r.  r   r   r  r   sequence_outputr   r1  r/   r/   r0   r=     s,   


z#Qwen3ForTokenClassification.forwardr  )rB   rC   rD   r$   r   r  r   r   r   r&   r   r   r   r   r   r   r=   rE   r/   r/   r-   r0   rG    sH    	
rG  c                       s   e Zd ZdZ fddZdd Zdd Zee									dd	e	e
j d
e	e
j de	e
j de	e de	e
j de	e
j de	e
j de	e de	e defddZ  ZS )Qwen3ForQuestionAnsweringtransformerc                    s2   t  | t|| _t|jd| _|   d S )Nr1   )	r#   r$   r   rM  r   rL   r+   
qa_outputsr   rR   r-   r/   r0   r$     s   
z"Qwen3ForQuestionAnswering.__init__c                 C   r!  rS   rM  r   r@   r/   r/   r0   r     r#  z.Qwen3ForQuestionAnswering.get_input_embeddingsc                 C   r$  rS   rO  r  r/   r/   r0   r    r%  z.Qwen3ForQuestionAnswering.set_input_embeddingsNr  rs   r`   r   r  start_positionsend_positionsr   r  rf   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)rs   r`   r   r  r   r  r   r2   rU   )r1  start_logits
end_logitsr:   r  )
rM  r  rN  splitsqueezer~   r3  r   r:   r  )r*   r  rs   r`   r   r  rP  rQ  r   r  r   r   rK  r   rR  rS  r1  r/   r/   r0   r=     s0   

z!Qwen3ForQuestionAnswering.forwardr  )rB   rC   rD   r   r$   r   r  r   r   r   r&   r   r   r   r   r   r   r=   rE   r/   r/   r-   r0   rL    sJ    	
rL  )r  rL  r   r   r:  rG  )Nr   )rn   )Ctypingr   r   r   r&   r   activationsr   cache_utilsr   r	   
generationr
   integrationsr   masking_utilsr   r   modeling_flash_attention_utilsr   modeling_layersr   modeling_outputsr   r   r   r   r   modeling_rope_utilsr   r   modeling_utilsr   r   processing_utilsr   utilsr   r   r   r   configuration_qwen3r   
get_loggerrB   r  Moduler!   rF   rZ   rd   r   r   rm   r   r   r   r   r   r   r   r  r  r:  rG  rL  __all__r/   r/   r/   r0   <module>   sv   


J6" 	lVF>