§
    9²¸jT‘  ã                   óü  — d dl Z d dlmZ d dlmZ d dlmZ d dlZd dlm	c 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 ddlmZmZmZ ddlmZmZ ddlm Z  ddl!m"Z"m#Z#m$Z$ 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/m0Z0m1Z1m2Z2 ddl3m4Z4m5Z5 ddl6m7Z7m8Z8 ddl9m:Z: ddl;m<Z< ddl=m>Z> ddl?m@Z@ ddlAmBZB ddlCmDZD ddlEmFZFmGZG ddlHmIZImJZJ ddlKmLZL d d!lMmNZN d"eOd#eOd$eOdz  d%edz  d&ePeOeOf         f
d'„ZQ G d(„ d)e+d*¬+¦  «        ZRe0 G d,„ d-e¦  «        ¦   «         ZS G d.„ d/e>¦  «        ZT G d0„ d1e	jU        ¦  «        ZV G d2„ d3e	jU        ¦  «        ZW G d4„ d5eB¦  «        ZX G d6„ d7e	jU        ¦  «        ZY G d8„ d9e	jU        ¦  «        ZZe1 G d:„ d;eJ¦  «        ¦   «         Z[ G d<„ d=eL¦  «        Z\ G d>„ d?e ¦  «        Z] G d@„ dAeI¦  «        Z^ e0dB¬C¦  «         G dD„ dEe^¦  «        ¦   «         Z_ e0dF¬C¦  «         G dG„ dHe^¦  «        ¦   «         Z` e0dI¬C¦  «        e G dJ„ dKe"¦  «        ¦   «         ¦   «         Za G dL„ dMe	jU        ¦  «        Zb G dN„ dOe	jU        ¦  «        Zc e0dP¬C¦  «         G dQ„ dRe^¦  «        ¦   «         Zdg dS¢ZedS )Té    N)ÚCallable)Ú	dataclass)Ú	Annotated)Únn)Ú
functionalé   )Úinitialization)ÚTorchvisionBackend)ÚBatchFeature)Úgroup_images_by_shapeÚreorder_images)Ú
ImageInputÚPILImageResamplingÚSizeDict)Úcreate_bidirectional_maskÚ(create_bidirectional_sliding_window_mask)ÚGradientCheckpointingLayer)ÚBaseModelOutputÚBaseModelOutputWithPoolingÚMaskedLMOutput)ÚROPE_INIT_FUNCTIONS)ÚALL_ATTENTION_FUNCTIONSÚPreTrainedModel)ÚImagesKwargsÚUnpack)Ú
TensorTypeÚTransformersKwargsÚauto_docstringÚno_inherit_decoratorÚtorch_compilable_check)ÚIMAGENET_STANDARD_MEANÚIMAGENET_STANDARD_STD)Úcan_return_tupleÚmaybe_autocast)Úcapture_outputs)Úpositive_inté   )ÚGemma4RMSNorm)Úapply_rotary_pos_emb)ÚLagunaRotaryEmbedding)Úconvert_image_to_patches)Úeager_attention_forwardÚ	repeat_kv)ÚMuseGlimmerPreTrainedModelÚMuseGlimmerTextAttention)ÚNemotronMLPé   )ÚNeoMMEConfigÚheightÚwidthÚmax_sideÚsizeÚreturnc                 óê  — |�|j         nd}|�|j        nd}|�| |z  |k     r|| |z  z  dz  nd}|�!t          ||t          | |¦  «        z  ¦  «        }|�t          ||| |z  z  dz  ¦  «        }|dk    r| |fS t          dt	          | |z  ¦  «        ¦  «        }t          dt	          ||z  ¦  «        ¦  «        }|durt          ||¦  «        |k    p|duo||z  |k    }	|	r†t          dt          j        | |z  ¦  «        ¦  «        }t          dt          j        ||z  ¦  «        ¦  «        }|�6||z  |k    r-||k    rt          d||z  ¦  «        }nt          d||z  ¦  «        }||fS |�‹| |z  |k     r‚||z  |k    r||fS t          dt          j        | |z  ¦  «        ¦  «        }
t          dt          j        ||z  ¦  «        ¦  «        }|durt          |
|¦  «        |k    p|duo|
|z  |k    }	|	s|
|fS ||fS )zHCompute integer height and width from the configured image size targets.Ng      à?ç      ð?r1   )Ú
min_pixelsÚ
max_pixelsÚminÚmaxÚroundÚmathÚfloorÚceil)r3   r4   r5   r6   r:   r;   ÚscaleÚresized_heightÚresized_widthÚexceeds_capÚceiled_heightÚceiled_widths               úy/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/neomme/modular_neomme.pyÚget_resize_output_sizerI   9   s€  € à$(Ð$4�”�¸$€JØ$(Ð$4�”�¸$€JØ6@Ð6LÐQWÐZ_ÑQ_ÐblÒQlÐQlˆZ˜6 E™>Ñ*¨sÑ2Ð2Ðru€EØÐÝ�E˜8¥c¨&°%Ñ&8Ô&8Ñ8Ñ9Ô9ˆØÐÝ�E˜J¨&°5©.Ñ9¸cÑAÑBÔBˆØ�‚|€|Ø�uˆ}Ðå˜�E &¨5¡.Ñ1Ô1Ñ2Ô2€NÝ˜�5 ¨¡Ñ/Ô/Ñ0Ô0€MØ 4Ð'ÐY­C°ÀÑ,NÔ,NÐQYÒ,Yð Ø˜$ÐÐN >°MÑ#AÀJÒ#Nð ð ð -Ý˜Q¥¤
¨6°E©>Ñ :Ô :Ñ;Ô;ˆÝ˜A�tœz¨%°%©-Ñ8Ô8Ñ9Ô9ˆØÐ! n°}Ñ&DÀzÒ&QÐ&QØ Ò.Ð.Ý!$ Q¨
°mÑ(CÑ!DÔ!D��å # A z°^Ñ'CÑ DÔ D�Ø˜}Ð,Ð,àÐ &¨5¡.°:Ò"=Ð"=Ø˜MÑ)¨ZÒ7Ð7Ø! =Ð0Ð0Ý˜A�tœy¨°%©Ñ8Ô8Ñ9Ô9ˆÝ˜1�dœi¨°©Ñ6Ô6Ñ7Ô7ˆØ tÐ+Ð[µ°MÀ<Ñ0PÔ0PÐS[Ò0[ð 
Ø˜dÐ"ÐP }°|Ñ'CÀjÒ'Pð 	ð ð 	/Ø  ,Ð.Ð.à˜=Ð(Ð(ó    c                   ón   — e Zd ZU dZee e¦   «         f         ed<   eedz   e¦   «         f         ed<   dS )ÚNeoMMEImageProcessorKwargsa  
    patch_size (`int`, *optional*, defaults to `self.patch_size`):
        Side, in pixels, of one patch token. The image is padded to a whole multiple of it.
    max_side (`int`, *optional*):
        Longest-side cap in pixels. Unset means no longest-side resize.
    Ú
patch_sizeNr5   )Ú__name__Ú
__module__Ú__qualname__Ú__doc__r   Úintr&   Ú__annotations__© rJ   rH   rL   rL   b   sZ   € € € € € € ðð ð ˜#˜|˜|™~œ~Ð-Ô.Ð.Ð.Ñ.Ø˜˜d™
 L L¡N¤NÐ2Ô3Ð3Ð3Ñ3Ð3Ð3rJ   rL   F)Útotalc                   ó¸  ‡ — e Zd ZdZeZej        Ze	Z
eZdZdZdZdZdZdZdZdZddgZdee         fˆ fd	„Zed
edee         defˆ fd„¦   «         Zdefˆ fd„Zdddedz  dedz  ddddf
d„Z dddededeef         fd„Z!d
e"d         de#dededz  dedz  ddde#de$de#de$e"e$         z  dz  de$e"e$         z  dz  de#dz  de%e&z  dz  defd„Z'd#d ed!edefd"„Z(ˆ xZ)S )$ÚNeoMMEImageProcessora�  
    Constructs an image processor for NeoMME.

    The processor converts each image to RGB and splits it into row-major patches. It preserves the original size by
    default, but `max_side` or a pixel-area `size` can resize images. It pads the bottom and right edges to complete
    the final patch. A batch returns one concatenated patch tensor and one grid height and width pair per image.
    Tgp?é    NÚpixel_valuesÚimage_grid_hwÚkwargsc                 ó:   •—  t          ¦   «         j        di |¤Ž d S )NrT   )ÚsuperÚ__init__)Úselfr[   Ú	__class__s     €rH   r^   zNeoMMEImageProcessor.__init__‰   s&   ø€ Ø�‰ŒÔÐ"Ð"˜6Ð"Ð"Ð"Ð"Ð"rJ   Úimagesr7   c                 ó8   •—  t          ¦   «         j        |fi |¤ŽS )a  
        Returns:
            [`BatchFeature`] with `pixel_values` of shape `(total_patches, 3 * patch_size ** 2)` and
            `image_grid_hw` of shape `(batch_size, 2)`. `pixel_values` concatenates patches from every image in the
            batch.
        )r]   Ú
preprocess)r_   ra   r[   r`   s      €rH   rc   zNeoMMEImageProcessor.preprocessŒ   s$   ø€ ð "�u‰wŒwÔ! &Ð3Ð3¨FÐ3Ð3Ð3rJ   c                 óè   •— |d         }|�1t          t          |¦  «        ¦  «        ddhk    rt          d¦  «        ‚|d         €|                     dd ¦  «          t	          ¦   «         j        di |¤ŽS )Nr6   r:   r;   z4size must contain exactly min_pixels and max_pixels.Ú	do_resizerT   )ÚsetÚdictÚ
ValueErrorÚpopr]   Ú_validate_preprocess_kwargs)r_   r[   r6   r`   s      €rH   rj   z0NeoMMEImageProcessor._validate_preprocess_kwargs–   sw   ø€ Ø�fŒ~ˆØÐ¥¥D¨¡J¤J¡¤°LÀ,Ð3OÒ OÐ OÝÐSÑTÔTÐTà�&Œ>Ð!à�JŠJ�{ DÑ)Ô)Ð)Ø2�u‰wŒwÔ2Ð<Ð<°VÐ<Ð<Ð<rJ   Úimageztorch.Tensorr5   r6   Úresamplez7PILImageResampling | tvF.InterpolationMode | int | Nonec                 óÌ   — |j         d         |j         d         }}t          ||||¦  «        \  }}||f||fk    r|S t          ||¬¦  «        }|                      |||d¬¦  «        S )Néþÿÿÿéÿÿÿÿ)r3   r4   T)rk   r6   rl   Ú	antialias)ÚshaperI   r   Úresize)	r_   rk   r5   r6   rl   r3   r4   rC   rD   s	            rH   Ú_resize_to_budgetz&NeoMMEImageProcessor._resize_to_budget    sx   € ð œ Bœ¨¬°R¬�ˆÝ(>¸vÀuÈhÐX\Ñ(]Ô(]Ñ%ˆ˜Ø˜MÐ*¨v°u¨oÒ=Ð=ØˆLÝ˜~°]ÐCÑCÔCˆØ�{Š{ ¨T¸HÐPTˆ{ÑUÔUÐUrJ   rM   c                 óº   — |j         d         |j         d         }}| |z   | |z   }}||z  |z
  }||z  |z
  }|s|rt          j        |dd||gd¬¦  «        }|||fS )Nrn   ro   r   )Úfill)rq   ÚtvFÚpad)	r_   rk   rM   r3   r4   Úgrid_heightÚ
grid_widthÚ
pad_heightÚ	pad_widths	            rH   Ú_pad_to_patch_gridz'NeoMMEImageProcessor._pad_to_patch_grid®   s�   € Øœ Bœ¨¬°R¬�ˆØ%+ G¨zÑ$9Ð":¸u¸fÈ
Ñ>RÐ<S�ZˆØ  :Ñ-°Ñ6ˆ
Ø Ñ+¨eÑ3ˆ	Øð 	J˜ð 	JÝ”G˜E A q¨)°ZÐ#@ÀqÐIÑIÔIˆEà�k :Ð-Ð-rJ   re   Ú
do_rescaleÚrescale_factorÚdo_normalizeÚ
image_meanÚ	image_stdÚdisable_groupingÚreturn_tensorsc           	      óÒ  — t          ||¬¦  «        \  }}i }|                     ¦   «         D ]$\  }}|r|                      ||||¦  «        }|||<   Œ%t          ||¦  «        }t          ||¬¦  «        \  }}i }i }|                     ¦   «         D ]‚\  }}|                      ||¦  «        \  }}}|                      ||||	|
|¦  «        }t          ||¦  «        ||<   t          j        ||ggt          |¦  «        z  t          j
        ¬¦  «        ||<   Œƒt          ||¦  «        }t          ||¦  «        }t          t          j        |d¬¦  «        t          j        |¦  «        dœ|¬¦  «        S )N)r‚   )Údtyper   ©Údim)rY   rZ   )ÚdataÚtensor_type)r   Úitemsrs   r   r|   Úrescale_and_normalizer+   ÚtorchÚtensorÚlenÚint64r   ÚcatÚstack)r_   ra   re   rM   r5   r6   rl   r}   r~   r   r€   r�   r‚   rƒ   r[   Úgrouped_imagesÚgrouped_images_indexÚresized_images_groupedrq   Ústacked_imagesÚresized_imagesÚprocessed_images_groupedÚimage_grids_groupedrx   ry   rY   rZ   s                              rH   Ú_preprocessz NeoMMEImageProcessor._preprocess¸   sË  € õ" 0EÀVÐ^nÐ/oÑ/oÔ/oÑ,ˆÐ,Ø!#ÐØ%3×%9Ò%9Ñ%;Ô%;ð 	;ð 	;Ñ!ˆE�>Øð bØ!%×!7Ò!7¸ÈÐRVÐX`Ñ!aÔ!a�Ø,:Ð" 5Ñ)Ð)Ý'Ð(>Ð@TÑUÔUˆå/DÀ^ÐfvÐ/wÑ/wÔ/wÑ,ˆÐ,Ø#%Ð Ø ÐØ%3×%9Ò%9Ñ%;Ô%;ð 
	ð 
	Ñ!ˆE�>ð 7;×6MÒ6MÈnÐ^hÑ6iÔ6iÑ3ˆN˜K¨Ø!×7Ò7Ø 
¨N¸LÈ*ÐV_ñô ˆNõ /GÀ~ÐWaÑ.bÔ.bÐ$ UÑ+Ý).¬Ø˜zÐ*Ð+­c°.Ñ.AÔ.AÑAÍÌð*ñ *ô *Ð Ñ&Ð&õ &Ð&>Ð@TÑUÔUˆÝ&Ð':Ð<PÑQÔQˆÝå %¤	¨,¸AÐ >Ñ >Ô >Ý!&¤¨]Ñ!;Ô!;ðð ð 'ð
ñ 
ô 
ð 	
rJ   r3   r4   c                 ó^  — |pi }|                      d| j        ¦  «        }|                      d| j        ¦  «        }|                      |                      d| j        ¦  «        ¬¦  «        d         }|                      d| j        ¦  «        rt          ||||¦  «        \  }}| |z   | |z   z  S )a  
        Return the number of image patches for one image, excluding row markers.

        The method applies the effective resize settings, then rounds each image dimension up to a whole patch.
        Values in `images_kwargs` override the processor settings.
        rM   r5   r6   )r6   re   )ÚgetrM   r5   Ú_standardize_kwargsr6   re   rI   )r_   r3   r4   Úimages_kwargsrM   r5   r6   s          rH   Úget_number_of_image_patchesz0NeoMMEImageProcessor.get_number_of_image_patchesê   s»   € ð &Ð+¨ˆØ"×&Ò& |°T´_ÑEÔEˆ
Ø ×$Ò$ Z°´Ñ?Ô?ˆØ×'Ò'¨]×->Ò->¸vÀtÄyÑ-QÔ-QÐ'ÑRÔRÐSYÔZˆà×Ò˜[¨$¬.Ñ9Ô9ð 	RÝ2°6¸5À(ÈDÑQÔQ‰MˆF�Eà�˜JÑ&Ð'¨e¨V°zÑ-AÐ+BÑCÐCrJ   ©N)*rN   rO   rP   rQ   rL   Úvalid_kwargsr   ÚBILINEARrl   r!   r€   r"   r�   Údo_convert_rgbre   r}   r~   r   rM   r5   r6   Úmodel_input_namesr   r^   r   r   r   rc   Útuplerj   rR   r   rs   r|   ÚlistÚboolÚfloatÚstrr   r™   rž   Ú__classcell__©r`   s   @rH   rW   rW   n   s«  ø€ € € € € ðð ð .€Là!Ô*€HØ'€JØ%€IØ€NØ€IØ€JØ€NØ€LØ€Jà€HØ€Dà'¨Ð9Ðð# Ð(BÔ!Cð #ð #ð #ð #ð #ð #ð ð4 ð 4°vÐ>XÔ7Yð 4Ð^jð 4ð 4ð 4ð 4ð 4ñ „^ð4ð=°uð =ð =ð =ð =ð =ð =ðVàðVð ˜‘*ðVð ˜‰oð	Vð
 LðVð 
ðVð Vð Vð Vð.¨ð .ÀCð .ÈEÐR`ÐbeÐgjÐRjÔLkð .ð .ð .ð .ð0
à�^Ô$ð0
ð ð0
ð ð	0
ð
 ˜‘*ð0
ð ˜‰oð0
ð Lð0
ð ð0
ð ð0
ð ð0
ð ˜D œKÑ'¨$Ñ.ð0
ð ˜4 œ;Ñ&¨Ñ-ð0
ð  ™+ð0
ð ˜jÑ(¨4Ñ/ð0
ð  
ð!0
ð 0
ð 0
ð 0
ðdDð D°#ð D¸cð DÐZ]ð Dð Dð Dð Dð Dð Dð Dð DrJ   rW   c                   ó   — e Zd ZdS )ÚNeoMMERMSNormN©rN   rO   rP   rT   rJ   rH   r¬   r¬   ü   ó   € € € € € Ø€DrJ   r¬   c                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Z	ˆ xZ
S )ÚNeoMMEEmbeddingszZFactorized (ALBERT-style) token embeddings: `vocab_size -> embedding_rank -> hidden_size`.Úconfigc                 óÜ   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          j        |j        |j        d¬¦  «        | _	        d S ©NF©Úbias)
r]   r^   r   Ú	EmbeddingÚ
vocab_sizeÚembedding_rankÚword_embeddingsÚLinearÚhidden_sizeÚembedding_projection©r_   r±   r`   s     €rH   r^   zNeoMMEEmbeddings.__init__  sY   ø€ Ý‰Œ×ÒÑÔÐÝ!œ|¨FÔ,=¸vÔ?TÑUÔUˆÔÝ$&¤I¨fÔ.CÀVÔEWÐ^cÐ$dÑ$dÔ$dˆÔ!Ð!Ð!rJ   Ú	input_idsr7   c                 óR   — |                       |                      |¦  «        ¦  «        S rŸ   )r¼   r¹   )r_   r¾   s     rH   ÚforwardzNeoMMEEmbeddings.forward  s$   € Ø×(Ò(¨×)=Ò)=¸iÑ)HÔ)HÑIÔIÐIrJ   )rN   rO   rP   rQ   r2   r^   rŒ   Ú
LongTensorÚTensorrÀ   r©   rª   s   @rH   r°   r°      s‚   ø€ € € € € ØdÐdðe˜|ð eð eð eð eð eð eð
J Ô!1ð J°e´lð Jð Jð Jð Jð Jð Jð Jð JrJ   r°   c                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Zˆ xZ	S )ÚNeoMMEPatchEmbeddingsz<Patch stem that maps flattened image patches to hidden size.r±   c                 óX  •— t          ¦   «                              ¦   «          t          j        |j        ¦  «        | _        t          j        |j        |j        dz  d¬¦  «        | _        t          j	        ¦   «         | _
        t          j        |j        dz  |j        d¬¦  «        | _        d S )Nr'   Fr´   T)r]   r^   r   Ú	LayerNormÚ	patch_dimÚnormrº   r»   Úup_projÚGELUÚact_fnÚ	down_projr½   s     €rH   r^   zNeoMMEPatchEmbeddings.__init__  s†   ø€ Ý‰Œ×ÒÑÔÐÝ”L Ô!1Ñ2Ô2ˆŒ	Ý”y Ô!1°6Ô3EÈÑ3IÐPUÐVÑVÔVˆŒÝ”g‘i”iˆŒÝœ 6Ô#5¸Ñ#9¸6Ô;MÐTXÐYÑYÔYˆŒˆˆrJ   rY   r7   c                 ó¢   — |                       |                      |¦  «        ¦  «        }|                      |                      |¦  «        ¦  «        S rŸ   )rÉ   rÈ   rÌ   rË   )r_   rY   Úhidden_statess      rH   rÀ   zNeoMMEPatchEmbeddings.forward  s=   € ØŸš T§Y¢Y¨|Ñ%<Ô%<Ñ=Ô=ˆØ�~Š~˜dŸkšk¨-Ñ8Ô8Ñ9Ô9Ð9rJ   )
rN   rO   rP   rQ   r2   r^   rŒ   rÂ   rÀ   r©   rª   s   @rH   rÄ   rÄ     sw   ø€ € € € € ØFÐFðZ˜|ð Zð Zð Zð Zð Zð Zð: E¤Lð :°U´\ð :ð :ð :ð :ð :ð :ð :ð :rJ   rÄ   c                   ór   — e Zd ZdZ	 d	dej        dej        dedz  deej        ej        f         fd„Z	d„ Z
dS )
ÚNeoMMERotaryEmbeddingzBTwo-axis interleaved M-RoPE with per-layer-type frequency spectra.NÚxÚposition_idsÚ
layer_typer7   c                 ó.  — t          | |› d�¦  «        }t          | |› d�¦  «        }|d d d d …d f                              ¦   «                              d|j        d         dd¦  «        }|d d …d d …d d d …f                              ¦   «         }t	          |j        j        t          ¦  «        r|j        j        dk    r|j        j        nd}t          |d¬	¦  «        5  ||z   	                    dd
¦  «        }	|	 
                    ¦   «         |z  }
|	                     ¦   «         |z  }d d d ¦  «         n# 1 swxY w Y   |                      |
¦  «        }
|                      |¦  «        }|
                     |j        ¦  «        |                     |j        ¦  «        fS )NÚ	_inv_freqÚ_attention_scalingr'   r1   ro   ÚmpsÚcpuF)Údevice_typeÚenabledr   )Úgetattrr§   Úexpandrq   Ú
isinstanceÚdeviceÚtyper¨   r$   Ú	transposeÚcosÚsinÚrecomposition_frequenciesÚtor…   )r_   rÑ   rÒ   rÓ   Úinv_freqÚattention_scalingÚinv_freq_expandedÚposition_ids_expandedrÙ   Úfreqsrá   râ   s               rH   rÀ   zNeoMMERotaryEmbedding.forward  sÇ  € õ ˜4 JÐ!9Ð!9Ð!9Ñ:Ô:ˆÝ# D¨ZÐ*KÐ*KÐ*KÑLÔLÐà$ T¨4°°°°DÐ%8Ô9×?Ò?ÑAÔA×HÒHÈÈLÔL^Ð_`ÔLaÐceÐghÑiÔiÐØ ,¨Q¨Q¨Q°°°°4¸¸¸¨]Ô ;× AÒ AÑ CÔ CÐå'1°!´(´-ÅÑ'EÔ'EÐkÈ!Ì(Ì-Ð[`ÒJ`ÐJ`�a”h”m�mÐfkˆÝ¨¸UÐCÑCÔCð 	2ð 	2à&Ð)>Ñ>×IÒIÈ!ÈQÑOÔOˆEØ—)’)‘+”+Ð 1Ñ1ˆCØ—)’)‘+”+Ð 1Ñ1ˆCð		2ð 	2ð 	2ñ 	2ô 	2ð 	2ð 	2ð 	2ð 	2ð 	2ð 	2øøøð 	2ð 	2ð 	2ð 	2ð ×,Ò,¨SÑ1Ô1ˆØ×,Ò,¨SÑ1Ô1ˆØ�vŠv�a”g‰Œ §¢ q¤w¡¤Ð/Ð/s   ÃAD.Ä.D2Ä5D2c                 óÔ   — |d         dddd…f         |d         dddd…f         }}t          j        ||gd¬¦  «                             d¦  «        }t          j        ||gd¬¦  «        S )	z]
        Recompose the frequencies into the final spatial layout used per each grid.
        r   .Nr'   r1   ro   r†   rn   )rŒ   r‘   Úflattenr�   )r_   ÚfreqÚfreq_rowÚfreq_colÚangless        rH   rã   z/NeoMMERotaryEmbedding.recomposition_frequencies2  st   € ð " !œW S¨!¨$¨Q¨$ YÔ/°°a´¸¸a¸dÀ¸d¸Ô1C�(ˆÝ”˜h¨Ð1°rÐ:Ñ:Ô:×BÒBÀ2ÑFÔFˆÝŒy˜& &Ð)¨rÐ2Ñ2Ô2Ð2rJ   rŸ   )rN   rO   rP   rQ   rŒ   rÂ   rÁ   r¨   r¤   rÀ   rã   rT   rJ   rH   rÐ   rÐ     s�   € € € € € ØLÐLð Y]ð0ð 0Ø”ð0Ø-2Ô-=ð0ØKNÐQUÉ:ð0à	ˆuŒ|˜Uœ\Ð)Ô	*ð0ð 0ð 0ð 0ð(3ð 3ð 3ð 3ð 3rJ   rÐ   c                   óV   ‡ — e Zd Zdefˆ fd„Zdej        dej        dej        fd„Zˆ xZS )ÚNeoMMEExclusiveSelfAttentionr±   c                 óÐ   •— t          ¦   «                              ¦   «          |j        |j        z  | _        t          j        t          j        |j        ¦  «        ¦  «        | _	        d S rŸ   )
r]   r^   Únum_attention_headsÚnum_key_value_headsÚnum_key_value_groupsr   Ú	ParameterrŒ   ÚzerosÚalphar½   s     €rH   r^   z%NeoMMEExclusiveSelfAttention.__init__>  sN   ø€ Ý‰Œ×ÒÑÔÐØ$*Ô$>À&ÔB\Ñ$\ˆÔ!Ý”\¥%¤+¨fÔ.HÑ"IÔ"IÑJÔJˆŒ
ˆ
ˆ
rJ   Úattn_outputÚvalue_statesr7   c                 ó¬  — t          || j        ¦  «                             dd¦  «        }t          j        |                     ¦   «         d¬¦  «                             |j        ¦  «        }||z                       dd¬¦  «        }t          j
        | j        ¦  «                             |j        ¦  «                             dddd¦  «        }|||z  |z  z
  S )Nr1   r'   ro   r†   T)Úkeepdim)r-   rõ   rà   ÚFÚ	normalizer§   rä   r…   ÚsumrŒ   Útanhrø   Úview)r_   rù   rú   Ú
value_unitÚ
projectionrB   s         rH   rÀ   z$NeoMMEExclusiveSelfAttention.forwardC  s¿   € Ý  ¨tÔ/HÑIÔI×SÒSÐTUÐWXÑYÔYˆÝ”[ ×!3Ò!3Ñ!5Ô!5¸2Ð>Ñ>Ô>×AÒAÀ+ÔBSÑTÔTˆ
Ø! JÑ.×3Ò3°BÀÐ3ÑEÔEˆ
Ý”
˜4œ:Ñ&Ô&×)Ò)¨+Ô*;Ñ<Ô<×AÒAÀ!ÀQÈÈAÑNÔNˆØ˜e jÑ0°JÑ>Ñ>Ð>rJ   ©	rN   rO   rP   r2   r^   rŒ   rÂ   rÀ   r©   rª   s   @rH   rñ   rñ   =  s|   ø€ € € € € ðK˜|ð Kð Kð Kð Kð Kð Kð
? 5¤<ð ?¸u¼|ð ?ÐPUÔP\ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?rJ   rñ   c                   óV   ‡ — e Zd Zdefˆ fd„Zdej        dej        dej        fd„Zˆ xZS )ÚNeoMMESigmoidGatedProjectionr±   c                 óô   •— t          ¦   «                              ¦   «          |j        |j        z  }t	          j        |j        |d¬¦  «        | _        t	          j        ||j        |j        ¬¦  «        | _	        d S r³   )
r]   r^   ró   Úhead_dimr   rº   r»   Ú	gate_projÚattention_biasÚo_proj)r_   r±   Úprojection_sizer`   s      €rH   r^   z%NeoMMESigmoidGatedProjection.__init__L  se   ø€ Ý‰Œ×ÒÑÔÐØ Ô4°v´ÑFˆÝœ 6Ô#5°ÈUÐSÑSÔSˆŒÝ”i °Ô1CÈ&ÔJ_Ð`Ñ`Ô`ˆŒˆˆrJ   rù   rÎ   r7   c                 ó|   — |                       |t          j        |                      |¦  «        ¦  «        z  ¦  «        S rŸ   )r  rŒ   Úsigmoidr	  )r_   rù   rÎ   s      rH   rÀ   z$NeoMMESigmoidGatedProjection.forwardR  s/   € Ø�{Š{˜;­¬°t·~²~ÀmÑ7TÔ7TÑ)UÔ)UÑUÑVÔVÐVrJ   r  rª   s   @rH   r  r  K  s‡   ø€ € € € € ða˜|ð að að að að að aðW 5¤<ð WÀÄð WÐQVÔQ]ð Wð Wð Wð Wð Wð Wð Wð WrJ   r  c                   óÖ   ‡ — e Zd ZdZdedefˆ fd„Z	 	 ddej        de	ej        ej        f         dej        dz  d	ej        dz  d
e
e         de	ej        ej        dz  f         fd„Zˆ xZS )ÚNeoMMEAttentionzõBidirectional grouped-query attention with QK-norm, M-RoPE, and a sigmoid output gate.

    QK-norm runs before rotary embedding, value embeddings are added after rotation, and
    exclusive self-attention is applied before the output gate.
    r±   Ú	layer_idxc                 ó¼  •— t          ¦   «                              ¦   «          | `| `| `| `|j        |         | _        |j        | _        d| _	        t          |j        |j        d¬¦  «        | _        t          |j        |j        d¬¦  «        | _        | j        r|j        |         j        dz   nd | _        t%          |¦  «        | _        t)          |¦  «        | _        d S )NF©Ú
with_scaler1   )r]   r^   Úqk_normÚqk_scale_factorr	  r  Úlayer_typesÚattention_typeró   Ú	is_causalr¬   r  Únorm_epsÚq_normÚk_normÚis_local_attentionÚper_layer_configÚsliding_windowrñ   Úexclusive_self_attentionr  Úoutput_projection©r_   r±   r  r`   s      €rH   r^   zNeoMMEAttention.__init__^  sÜ   ø€ Ý‰Œ×ÒÑÔÐØˆLØÐ ØˆNØˆKà$Ô0°Ô;ˆÔØ#)Ô#=ˆÔ ØˆŒÝ# F¤O°V´_ÐQVÐWÑWÔWˆŒÝ# F¤O°V´_ÐQVÐWÑWÔWˆŒð
 FJÔE\ÐfˆFÔ# IÔ.Ô=ÀÑAÐAÐbfð 	Ôõ )EÀVÑ(LÔ(LˆÔ%Ý!=¸fÑ!EÔ!EˆÔÐÐrJ   NrÎ   Úposition_embeddingsÚattention_maskÚvalue_embedsr[   r7   c                 ó¦  — |j         d d…         }g |¢d‘| j        ‘R }|                      |¦  «                             |¦  «                             dd¦  «        }|                      |¦  «                             |¦  «                             dd¦  «        }	|                      |¦  «                             |¦  «                             dd¦  «        }
|                      |¦  «        }|                      |	¦  «        }	|\  }}t          ||	||¦  «        \  }}	|�,|
|                     |¦  «                             dd¦  «        z   }
t          j        | j        j        t          ¦  «        } || ||	|
|f| j        r| j        nd| j        | j        dœ|¤Ž\  }}|                      ||
¦  «        } |j        g |¢d‘R Ž }|                      ||¦  «        }||fS )Nro   r1   r'   ç        )ÚdropoutÚscalingr  )rq   r  Úq_projr  rà   Úk_projÚv_projr  r  r)   r   Úget_interfacer±   Ú_attn_implementationr,   ÚtrainingÚattention_dropoutr)  r  r   Úreshaper!  )r_   rÎ   r#  r$  r%  r[   Úinput_shapeÚhidden_shapeÚquery_statesÚ
key_statesrú   rá   râ   Úattention_interfacerù   Úattn_weightss                   rH   rÀ   zNeoMMEAttention.forwards  s   € ð $Ô)¨#¨2¨#Ô.ˆØ8˜Ð8 bÐ8¨$¬-Ð8Ð8ˆà—{’{ =Ñ1Ô1×6Ò6°|ÑDÔD×NÒNÈqÐRSÑTÔTˆØ—[’[ Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆ
Ø—{’{ =Ñ1Ô1×6Ò6°|ÑDÔD×NÒNÈqÐRSÑTÔTˆà—{’{ <Ñ0Ô0ˆØ—[’[ Ñ,Ô,ˆ
à&‰ˆˆSÝ#7¸ÀjÐRUÐWZÑ#[Ô#[Ñ ˆ�jØÐ#Ø'¨,×*;Ò*;¸LÑ*IÔ*I×*SÒ*SÐTUÐWXÑ*YÔ*YÑYˆLå(?Ô(MØŒKÔ,Õ.Eñ)
ô )
Ðð %8Ð$7ØØØØØð
%
ð /3¬mÐD�DÔ*Ð*ÀØ”LØÔ.ð
%
ð 
%
ð ð
%
ð 
%
Ñ!ˆ�\ð ×3Ò3°KÀÑNÔNˆØ)�kÔ)Ð;¨;Ð;¸Ð;Ð;Ð;ˆØ×,Ò,¨[¸-ÑHÔHˆØ˜LÐ(Ð(rJ   )NN©rN   rO   rP   rQ   r2   rR   r^   rŒ   rÂ   r¤   r   r   rÀ   r©   rª   s   @rH   r  r  V  sï   ø€ € € € € ðð ðF˜|ð F¸ð Fð Fð Fð Fð Fð Fð2 /3Ø,0ð*)ð *)à”|ð*)ð # 5¤<°´Ð#=Ô>ð*)ð œ tÑ+ð	*)ð
 ”l TÑ)ð*)ð Ð+Ô,ð*)ð 
ˆuŒ|˜Uœ\¨DÑ0Ð0Ô	1ð*)ð *)ð *)ð *)ð *)ð *)ð *)ð *)rJ   r  c                   ó   — e Zd ZdS )Ú	NeoMMEMLPNr­   rT   rJ   rH   r:  r:     r®   rJ   r:  c                   óÌ   ‡ — e Zd ZdZdedefˆ fd„Z	 	 	 ddej        dej        dej        dz  d	e	ej        ej        f         dz  d
ej        dz  de
e         dej        fd„Zˆ xZS )ÚNeoMMEEncoderLayerzGPre-norm encoder layer with initial-state mixing and muP depth scaling.r±   r  c                 ó´  •— t          ¦   «                              ¦   «          t          ||¦  «        | _        t	          |¦  «        | _        t          j        t          j	        ddg¦  «        ¦  «        | _
        t          |j        |j        d¬¦  «        | _        t          |j        |j        d¬¦  «        | _        |j        | _        |j        |         | _        d S )Nr9   r'  Fr  )r]   r^   r  Ú	self_attnr:  Úmlpr   rö   rŒ   r�   Úlambdasr¬   r»   r  Úinput_layernormÚpost_attention_layernormÚresidual_multiplierr  r  r"  s      €rH   r^   zNeoMMEEncoderLayer.__init__§  s³   ø€ Ý‰Œ×ÒÑÔÐÝ(¨°Ñ;Ô;ˆŒÝ˜VÑ$Ô$ˆŒÝ”|¥E¤L°#°s°Ñ$<Ô$<Ñ=Ô=ˆŒÝ,¨VÔ-?ÀÄÐ]bÐcÑcÔcˆÔÝ(5°fÔ6HÈ&Ì/ÐfkÐ(lÑ(lÔ(lˆÔ%Ø#)Ô#=ˆÔ Ø$Ô0°Ô;ˆÔÐÐrJ   NrÎ   Úinitial_hidden_statesr%  r#  r$  r[   r7   c                 ó   — | j         d         |z  | j         d         |z  z   }|                      |¦  «        } | j        |f|||dœ|¤Ž\  }	}
|| j        |	z  z   }|                      |¦  «        }|                      |¦  «        }|| j        |z  z   S )Nr   r1   )r#  r$  r%  )r@  rA  r>  rC  rB  r?  )r_   rÎ   rD  r%  r#  r$  r[   Úmixed_statesÚnormed_statesrù   Ú_Ú
mlp_outputs               rH   rÀ   zNeoMMEEncoderLayer.forward±  sº   € ð ”| A”¨Ñ6¸¼Àa¼ÐK`Ñ9`Ñ`ˆØ×,Ò,¨\Ñ:Ô:ˆØ'˜œØð
à 3Ø)Ø%ð	
ð 
ð
 ð
ð 
‰ˆ�Qð &¨Ô(@À;Ñ(NÑNˆà×5Ò5°mÑDÔDˆØ—X’X˜mÑ,Ô,ˆ
Ø˜tÔ7¸*ÑDÑDÐDrJ   ©NNNr8  rª   s   @rH   r<  r<  ¤  sö   ø€ € € € € ØQÐQð<˜|ð <¸ð <ð <ð <ð <ð <ð <ð -1ØHLØ.2ðEð Eà”|ðEð  %œ|ðEð ”l TÑ)ð	Eð
 # 5¤<°´Ð#=Ô>ÀÑEðEð œ tÑ+ðEð Ð+Ô,ðEð 
ŒðEð Eð Eð Eð Eð Eð Eð ErJ   r<  c            	       óÈ   ‡ — e Zd ZdZdgZeedœZdej	        fd„Z
dej	        ddfd„Z ej        ¦   «         d	ej        fd
„¦   «         Z	 ddededz  dedej	        fˆ fd„Zˆ xZS )ÚNeoMMEPreTrainedModel)rk   Útextr<  )rÎ   Ú
attentionsr7   c                 óF   — t          | | j        | ¦  «        }|j        j        S rŸ   ©rÛ   Úbase_model_prefixÚembed_tokensr¹   )r_   Úbackbones     rH   Úget_input_embeddingsz*NeoMMEPreTrainedModel.get_input_embeddingsÒ  s"   € Ý˜4 Ô!7¸Ñ>Ô>ˆØÔ$Ô4Ð4rJ   ÚvalueNc                 óJ   — t          | | j        | ¦  «        }||j        _        d S rŸ   rP  )r_   rU  rS  s      rH   Úset_input_embeddingsz*NeoMMEPreTrainedModel.set_input_embeddingsÖ  s'   € Ý˜4 Ô!7¸Ñ>Ô>ˆØ05ˆÔÔ-Ð-Ð-rJ   Úmodulec                 óÐ  — t          j        | |¦  «         t          |t          ¦  «        r0t	          j        |j        j        d| j        j	        dz  ¬¦  «         d S t          |t          ¦  «        rt	          j        |j        ¦  «         d S t          |t          ¦  «        r t	          j        |j        j        ¦  «         d S t          |t          ¦  «        r t	          j        |j        j        ¦  «         d S t          |t"          ¦  «        r0t	          j        |j        t)          j        ddg¦  «        ¦  «         d S t          |t,          ¦  «        r t	          j        |j        j        ¦  «         d S t          |t0          ¦  «        r0t	          j        |j        j        d| j        j	        dz  ¬¦  «         d S t          |t4          ¦  «        r›|j        D ]•}|j        }|j        |         dk    rt<          |j        |                  } ||j        |¬¦  «        \  }}t	          j        t?          ||› d�¦  «        |¦  «         t	          j        t?          ||› d�¦  «        |¦  «         Œ”d S d S )	Nr'  g      à¿)ÚmeanÚstdr9   Údefault)rÓ   rÕ   Ú_original_inv_freq) r   Ú_init_weightsrÝ   r°   ÚinitÚnormal_r¹   Úweightr±   r¸   rñ   Úzeros_rø   r  r  r:  rÌ   r<  Úcopy_r@  rŒ   r�   ÚNeoMMEModelÚvalue_embeddingsÚNeoMMEForMaskedLMÚlm_headrÐ   r  Úcompute_default_rope_parametersÚ	rope_typer   rÛ   )r_   rX  rÓ   Úrope_init_fnrå   rH  s         rH   r^  z#NeoMMEPreTrainedModel._init_weightsÚ  sc  € õ 	Ô% d¨FÑ3Ô3Ð3å�fÕ.Ñ/Ô/ð 	YÝŒL˜Ô/Ô6¸SÀdÄkÔF`ÐbfÑFfÐgÑgÔgÐgÐgÐgÝ˜Õ <Ñ=Ô=ð 	YÝŒK˜œÑ%Ô%Ð%Ð%Ð%Ý˜Õ <Ñ=Ô=ð 	YåŒK˜œÔ,Ñ-Ô-Ð-Ð-Ð-Ý˜¥	Ñ*Ô*ð 	YÝŒK˜Ô(Ô/Ñ0Ô0Ð0Ð0Ð0Ý˜Õ 2Ñ3Ô3ð 	YÝŒJ�v”~¥u¤|°S¸#°JÑ'?Ô'?Ñ@Ô@Ð@Ð@Ð@Ý˜¥Ñ,Ô,ð 	YÝŒK˜Ô/Ô6Ñ7Ô7Ð7Ð7Ð7Ý˜Õ 1Ñ2Ô2ð 		YÝŒL˜œÔ.°S¸d¼kÔ>XÐZ^Ñ>^Ð_Ñ_Ô_Ð_Ð_Ð_Ý˜Õ 5Ñ6Ô6ð 	YØ$Ô0ð Yð Y�
Ø%ÔE�ØÔ# JÔ/°9Ò<Ð<Ý#6°vÔ7GÈ
Ô7SÔ#T�LØ*˜l¨6¬=ÀZÐPÑPÔP‘�˜!Ý”
�7 6¨jÐ+CÐ+CÐ+CÑDÔDÀhÑOÔOÐOÝ”
�7 6¨jÐ+LÐ+LÐ+LÑMÔMÈxÑXÔXÐXÐXð	Yð 	YðYð YrJ   TÚnew_num_tokensÚpad_to_multiple_ofÚmean_resizingc                 ó>  •— t          ¦   «                              |||¦  «        }t          | | j        | ¦  «        }|                      |j        |j        j        d         |¬¦  «        }t          j	        |j
        |j        ¦  «        |_        |j        |j        _        |S )z0Resize word and value embedding tables together.r   )rm  )r]   Ú_resize_token_embeddingsrÛ   rQ  Ú_get_resized_embeddingsre  ra  rq   r   r¶   Únum_embeddingsÚembedding_dim)r_   rk  rl  rm  r¹   rS  Úresizedr`   s          €rH   ro  z.NeoMMEPreTrainedModel._resize_token_embeddings÷  s•   ø€ õ  ™'œ'×:Ò:¸>ÐK]Ð_lÑmÔmˆÝ˜4 Ô!7¸Ñ>Ô>ˆØ×.Ò.ØÔ% Ô'=Ô'CÀAÔ'FÐVcð /ñ 
ô 
ˆõ %'¤L°Ô1GÈÔI^Ñ$_Ô$_ˆÔ!Ø+2¬>ˆÔ!Ô(ØÐrJ   )NT)rN   rO   rP   Úinput_modalitiesÚ_no_split_modulesr<  r  Ú_can_record_outputsr   r¶   rT  rW  rŒ   Úno_gradÚModuler^  rR   r¦   ro  r©   rª   s   @rH   rL  rL  Ê  s  ø€ € € € € Ø(ÐØ-Ð.Ðà+Ø%ðð Ðð
5 b¤lð 5ð 5ð 5ð 5ð6¨"¬,ð 6¸4ð 6ð 6ð 6ð 6ð €U„]�_„_ðY B¤Ið Yð Yð Yñ „_ðYð: aeðð Ø!ðØ7:¸T±zðØY]ðà	Œðð ð ð ð ð ð ð ð ð rJ   rL  zu
    The bare NeoMME model. It encodes text tokens and image patches with one bidirectional Transformer encoder.
    ©Úcustom_introc                   óB  ‡ — e Zd Zdefˆ fd„Ze ed¬¦  «        dej        de	e
         defd„¦   «         ¦   «         Zd	ej        d
ej        dej        fd„Zee	 	 	 dd	ej        dej        dz  dej        dz  dej        dz  de	e
         defd„¦   «         ¦   «         Zˆ xZS )rd  r±   c                 óª  •‡— t          ¦   «                              ‰¦  «         t          ‰¦  «        | _        t	          ‰¦  «        | _        t          ‰¦  «        | _        t          ‰j	        ‰j
        d¬¦  «        | _        t          ‰j	        ‰j
        d¬¦  «        | _        t          j        ˆfd„t          ‰j        ¦  «        D ¦   «         ¦  «        | _        d„ t%          ‰j        ¦  «        D ¦   «         }t          j        ‰j        ‰j        ‰j        z  ¦  «        | _        |d         |d         h| _        d| _        |                      ¦   «          d S )NFr  c                 ó0   •— g | ]}t          ‰|¦  «        ‘ŒS rT   )r<  )Ú.0r  r±   s     €rH   ú
<listcomp>z(NeoMMEModel.__init__.<locals>.<listcomp>  s$   ø€ ÐdÐdÐd°yÕ ¨	Ñ2Ô2ÐdÐdÐdrJ   c                 ó$   — g | ]\  }}|d k    ¯|‘ŒS )Úfull_attentionrT   )r~  ÚirÓ   s      rH   r  z(NeoMMEModel.__init__.<locals>.<listcomp>  s(   € ÐpÐpÐp™}˜q *ÐQ[Ð_oÒQoÐQo˜ÐQoÐQoÐQorJ   r   ro   )r]   r^   r°   rR  rÄ   Úpatch_embeddingsrÐ   Ú
rotary_embr¬   r»   r  Úembedding_normÚ
final_normr   Ú
ModuleListÚrangeÚnum_hidden_layersÚlayersÚ	enumerater  r¶   r·   rô   r  re  Úvalue_embedding_layersÚgradient_checkpointingÚ	post_init)r_   r±   Úglobal_layersr`   s    ` €rH   r^   zNeoMMEModel.__init__  s2  øø€ Ý‰Œ×Ò˜Ñ Ô Ð Ý,¨VÑ4Ô4ˆÔÝ 5°fÑ =Ô =ˆÔÝ/°Ñ7Ô7ˆŒÝ+¨FÔ,>ÀÄÐ\aÐbÑbÔbˆÔÝ'¨Ô(:¸F¼OÐX]Ð^Ñ^Ô^ˆŒÝ”mØdÐdÐdÐdÅEÈ&ÔJbÑDcÔDcÐdÑdÔdñ
ô 
ˆŒð qÐpµ	¸&Ô:LÑ0MÔ0MÐpÑpÔpˆÝ "¤¨VÔ->ÀÔ@ZÐ]cÔ]lÑ@lÑ mÔ mˆÔØ'4°QÔ'7¸ÀrÔ9JÐ&KˆÔ#Ø&+ˆÔ#Ø�ŠÑÔÐÐÐrJ   z=Projects flattened image patches into the model hidden space.ry  rY   r[   r7   c           	      óB  — |j         d         | j        j        k    r8t          d|j         d         › d| j        j        › d| j        j        › d�¦  «        ‚|                      |                     | j        j        j        j	        ¦  «        ¦  «        }t          ||¬¦  «        S )Nro   zpixel_values has patch width z but the model expects z( (= 3 * patch_size ** 2 with patch_size=ú))Úlast_hidden_stateÚpooler_output)rq   r±   rÇ   rh   rM   rƒ  rä   rÈ   ra  r…   r   )r_   rY   r[   Úimage_featuress       rH   Úget_image_featureszNeoMMEModel.get_image_features  sº   € ð
 Ô˜bÔ! T¤[Ô%:Ò:Ð:Ýðl°Ô0BÀ2Ô0Fð lð lØ”;Ô(ðlð lØRVÔR]ÔRhðlð lð lñô ð ð ×.Ò.¨|¯ª¸tÔ?TÔ?YÔ?`Ô?fÑ/gÔ/gÑhÔhˆÝ)¸NÐZhÐiÑiÔiÐirJ   r¾   r”  c                 ó~  ‡‡— t          j        |dd…dd…f         d| j        j        pd¬¦  «        }|| j        j        k    || j        j        k    z  }|                     ¦   «         Št          ‰‰j        d         k    ˆˆfd„¦  «         | 	                    d¦  «         
                    dd‰j        d         ¦  «        S )zPFind patch placeholders and validate that every image feature has a destination.Nro   )r1   r   r   )rU  c                  óF   •— d‰ j         d         › dt          ‰¦  «        › d�S )NzGot r   z image patches for z image placeholder tokens)rq   rR   )r”  Únum_image_tokenss   €€rH   ú<lambda>z2NeoMMEModel.get_placeholder_mask.<locals>.<lambda>1  s,   ø€ Ðw˜>Ô/°Ô2ÐwÐwÅsÐK[ÑG\ÔG\ÐwÐwÐw€ rJ   )rý   rw   r±   Úpad_token_idÚimage_token_idÚdocument_token_idrÿ   r    rq   Ú	unsqueezerÜ   )r_   r¾   r”  Úprevious_idsÚ
image_maskr˜  s     `  @rH   Úget_placeholder_maskz NeoMMEModel.get_placeholder_mask)  sÇ   øø€ å”u˜Y q q q¨#¨2¨# vÔ.°¸d¼kÔ>VÐ>[ÐZ[Ð\Ñ\Ô\ˆØ 4¤;Ô#=Ò=À,ÐRVÔR]ÔRoÒBoÑpˆ
à%Ÿ>š>Ñ+Ô+ÐÝØ Ô 4°QÔ 7Ò7ØwÐwÐwÐwÐwñ	
ô 	
ð 	
ð ×#Ò# BÑ'Ô'×.Ò.¨r°2°~Ô7KÈBÔ7OÑPÔPÐPrJ   Nr$  rÒ   c                 ó$  ‡ ‡‡— ‰                       |¦  «        Š|�M‰                      |d¬¦  «        }‰                      ||j        ¦  «        }‰                     ||j        ¦  «        Š‰j        dd…         \  }}	‰€/t          j        |	‰j        ¬¦  «         	                    |d¦  «        Š‰j
        dk    r*‰                     d¦  «         	                    ddd¦  «        Š‰                      ‰¦  «        xŠ}
t          |x}t          ¦  «        s_i }‰|dœ}t          ‰ j        j        ¦  «        D ]>}‰ j        j        |         }|j        �t)          dd	|i|¤Ž||<   Œ-t+          dd	|i|¤Ž||<   Œ?ˆˆˆ fd
„t-          ‰ j        j        ¦  «        D ¦   «         }‰                      |¦  «        }t3          ‰ j        ¦  «        D ]/\  }} |‰|
|‰ j        v r|ndf||j                 ||         dœ|¤ŽŠŒ0‰                      ‰¦  «        Št=          ‰¬¦  «        S )a6  
        position_ids (`torch.LongTensor` of shape `(2, batch_size, sequence_length)` or `(batch_size, sequence_length)`, *optional*):
            Positions for the input tokens. [`NeoMMEProcessor`] returns two-axis positions for document images. A
            one-axis position tensor is used for text inputs.
        pixel_values (`torch.Tensor` of shape `(num_patches, 3 * patch_size ** 2)`, *optional*):
            Flattened image patches returned by [`NeoMMEProcessor`]. The model places these patches at image
            placeholders in `input_ids`.
        NT)Úreturn_dictr'   )rÞ   ro   r   )Úinputs_embedsr$  r±   c                 ó@   •— i | ]}|‰                      ‰‰|¦  «        “ŒS rT   )r„  )r~  rÓ   rÎ   rÒ   r_   s     €€€rH   ú
<dictcomp>z'NeoMMEModel.forward.<locals>.<dictcomp>g  s;   ø€ ð 
ð 
ð 
àð ˜Ÿš¨°|ÀZÑPÔPð
ð 
ð 
rJ   )r#  r$  )r’  rT   )rR  r•  r   r“  Úmasked_scatterrq   rŒ   ÚarangerÞ   rÜ   Úndimr�  r…  rÝ   rg   rˆ  r±   r‰  r  r  r   r   rf   r  re  r‹  rŠ  rŒ  r  r†  r   )r_   r¾   r$  rÒ   rY   r[   Úimage_outputsrŸ  Ú
batch_sizeÚseq_lenrD  Úattention_mask_mappingÚmask_kwargsÚlayer_idr  r#  r%  r  Úencoder_layerrÎ   s   `  `               @rH   rÀ   zNeoMMEModel.forward5  s¥  øøø€ ð$ ×)Ò)¨)Ñ4Ô4ˆØÐ#Ø ×3Ò3°LÈdÐ3ÑSÔSˆMØ×2Ò2°9¸mÔ>YÑZÔZˆJØ)×8Ò8¸À]ÔE`ÑaÔaˆMà+Ô1°"°1°"Ô5Ñˆ
�GàÐÝ œ<¨¸Ô8LÐMÑMÔM×TÒTÐU_ÐacÑdÔdˆLØÔ Ò!Ð!Ø'×1Ò1°!Ñ4Ô4×;Ò;¸A¸rÀ2ÑFÔFˆLð 15×0CÒ0CÀMÑ0RÔ0RÐRˆÐ-å°NÐBÐ0ÅDÑIÔIð 	ØEGÐ"Ø,9È^Ð\Ð\ˆKÝ! $¤+Ô"?Ñ@Ô@ð ð �Ø#'¤;Ô#?ÀÔ#IÐ Ø#Ô2Ð>Ý7_ð 8ð 8Ø/ð8à%ð8ð 8Ð*¨8Ñ4Ð4õ
 8Qð 8ð 8Ø/ð8à%ð8ð 8Ð*¨8Ñ4Ð4ð

ð 
ð 
ð 
ð 
ð 
å! $¤+Ô"9Ñ:Ô:ð
ñ 
ô 
Ðð
 ×,Ò,¨YÑ7Ô7ˆå(1°$´+Ñ(>Ô(>ð 		ð 		Ñ$ˆI�}à)˜MØØ%Ø )¨TÔ-HÐ HÐ H��Èdðð %8¸Ô8TÔ$UØ5°iÔ@ðð ð ðð ˆMˆMð Ÿš¨Ñ6Ô6ˆÝ°Ð?Ñ?Ô?Ð?rJ   rJ  )rN   rO   rP   r2   r^   r#   r   rŒ   rÂ   r   r   r   r•  rÁ   Ú
BoolTensorr   r%   r   rÀ   r©   rª   s   @rH   rd  rd    s’  ø€ € € € € ð˜|ð ð ð ð ð ð ð" Ø€^Ð!`ÐaÑaÔað	jØ!œLð	jØ4:Ð;MÔ4Nð	jà	#ð	jð 	jð 	jñ bÔañ Ôð	jð
Q¨eÔ.>ð 
QÐPUÔP\ð 
QÐafÔaqð 
Qð 
Qð 
Qð 
Qð Øð /3Ø04Ø,0ðC@ð C@àÔ#ðC@ð œ tÑ+ðC@ð Ô&¨Ñ-ð	C@ð
 ”l TÑ)ðC@ð Ð+Ô,ðC@ð 
ðC@ð C@ð C@ñ „^ñ „_ðC@ð C@ð C@ð C@ð C@rJ   rd  zB
    The NeoMME model with a factorized masked token decoder.
    c                   óÚ   ‡ — e Zd ZdddœZdefˆ fd„Zee	 	 	 	 	 ddej	        dz  dej
        dz  d	ej	        dz  d
ej
        dz  dej	        dz  dee         defd„¦   «         ¦   «         Zˆ xZS )rf  z)model.embed_tokens.word_embeddings.weightz.model.embed_tokens.embedding_projection.weight)zlm_head.weightzunembedding_projection.weightr±   c                 ó2  •— t          ¦   «                              |¦  «         t          |¦  «        | _        t	          j        |j        |j        d¬¦  «        | _        t	          j        |j        |j	        d¬¦  «        | _
        |                      ¦   «          d S r³   )r]   r^   rd  Úmodelr   rº   r¸   r»   Úunembedding_projectionr·   rg  rŽ  r½   s     €rH   r^   zNeoMMEForMaskedLM.__init__ˆ  s   ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ý  Ñ(Ô(ˆŒ
Ý&(¤i°Ô0EÀvÔGYÐ`eÐ&fÑ&fÔ&fˆÔ#Ý”y Ô!6¸Ô8IÐPUÐVÑVÔVˆŒØ�ŠÑÔÐÐÐrJ   Nr¾   r$  rÒ   rY   Úlabelsr[   r7   c                 óô   —  | j         d||||dœ|¤Ž}|j        }|| j        j        z  }|                      |¦  «        }	d}
|� | j        |	|fd| j        j        i|¤Ž}
t          |
|	|j	        |j
        ¬¦  «        S )a&  
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for the masked-language-modeling loss. Indices should be in `[0, ..., config.vocab_size - 1]`
            or `-100`; only tokens with a label different from `-100` contribute.
        ©r¾   r$  rÒ   rY   Nr·   )ÚlossÚlogitsrÎ   rN  rT   )r³  r’  r´  ra  rg  Úloss_functionr±   r·   r   rÎ   rN  )r_   r¾   r$  rÒ   rY   rµ  r[   ÚoutputsrÎ   r¹  r¸  s              rH   rÀ   zNeoMMEForMaskedLM.forward�  s½   € ð  $. 4¤:ð $
ØØ)Ø%Ø%ð	$
ð $
ð
 ð$
ð $
ˆð  Ô1ˆØ%¨Ô(CÔ(JÑJˆØ—’˜mÑ,Ô,ˆàˆØÐØ%�4Ô% f¨fÐbÐbÀÄÔAWÐbÐ[aÐbÐbˆDåØØØ!Ô/ØÔ)ð	
ñ 
ô 
ð 	
rJ   )NNNNN)rN   rO   rP   Ú_tied_weights_keysr2   r^   r#   r   rŒ   rÁ   rÂ   r   r   r   rÀ   r©   rª   s   @rH   rf  rf  }  s  ø€ € € € € ð FØ)Yðð Ðð
˜|ð ð ð ð ð ð ð Øð .2Ø.2Ø04Ø,0Ø*.ð"
ð "
àÔ# dÑ*ð"
ð œ tÑ+ð"
ð Ô&¨Ñ-ð	"
ð
 ”l TÑ)ð"
ð Ô  4Ñ'ð"
ð Ð+Ô,ð"
ð 
ð"
ð "
ð "
ñ „^ñ Ôð"
ð "
ð "
ð "
ð "
rJ   rf  z1
    Output type for [`NeoMMEForRetrieval`].
    c                   ón   — e Zd ZU dZdZej        dz  ed<   dZej        dz  ed<   dZ	ej        dz  ed<   dS )ÚNeoMMEForRetrievalOutputau  
    loss (`torch.FloatTensor` of shape `(1,)`, *optional*):
        Retrieval loss. This value is always `None`.
    embeddings (`torch.FloatTensor` of shape `(batch_size, sequence_length, embedding_dim)`, *optional*):
        Normalized token embeddings for late-interaction retrieval. Padding rows are zeroed. Score them with MeanMaxSim.
    dense_embeddings (`torch.FloatTensor` of shape `(batch_size, hidden_size)` or `(batch_size, dense_dim)`, *optional*):
        A normalized mean-pooled embedding for each input. When `dense_dim` is set, the last dimension is `dense_dim`.
        Score them with cosine similarity.
    Nr¸  Ú
embeddingsÚdense_embeddings)
rN   rO   rP   rQ   r¸  rŒ   ÚFloatTensorrS   r¿  rÀ  rT   rJ   rH   r¾  r¾  ¶  sg   € € € € € € ðð ð &*€Dˆ%Ô
˜dÑ
"Ð)Ð)Ñ)Ø+/€J�Ô! DÑ(Ð/Ð/Ñ/Ø15Ð�eÔ'¨$Ñ.Ð5Ð5Ñ5Ð5Ð5rJ   r¾  c                   óV   ‡ — e Zd Zdefˆ fd„Zdej        dej        dej        fd„Zˆ xZS )ÚNeoMMEMultiVectorHeadr±   c                 ó”   •— t          ¦   «                              ¦   «          t          j        |j        |j        d¬¦  «        | _        d S r³   )r]   r^   r   rº   r»   rr  Úprojr½   s     €rH   r^   zNeoMMEMultiVectorHead.__init__Í  s;   ø€ Ý‰Œ×ÒÑÔÐÝ”I˜fÔ0°&Ô2FÈUÐSÑSÔSˆŒ	ˆ	ˆ	rJ   rÎ   r$  r7   c                 ó  — |                       |                     | j         j        j        ¦  «        ¦  «        }t	          j        |d¬¦  «        }|                     |                     ¦   «                              d¦  «         d¦  «        S )Nro   r†   r'  )	rÅ  rä   ra  r…   rý   rþ   Úmasked_fillr¦   r�  )r_   rÎ   r$  r¿  s       rH   rÀ   zNeoMMEMultiVectorHead.forwardÑ  sp   € Ø—Y’Y˜}×/Ò/°´	Ô0@Ô0FÑGÔGÑHÔHˆ
Ý”[ °Ð4Ñ4Ô4ˆ
à×%Ò% ~×':Ò':Ñ'<Ô'<×'FÒ'FÀrÑ'JÔ'JÐ&JÈCÑPÔPÐPrJ   r  rª   s   @rH   rÃ  rÃ  Ì  s‡   ø€ € € € € ðT˜|ð Tð Tð Tð Tð Tð TðQ U¤\ð QÀ5Ä<ð QÐTYÔT`ð Qð Qð Qð Qð Qð Qð Qð QrJ   rÃ  c            	       óN   — e Zd Z	 ddej        dej        dedz  dej        fd„ZdS )ÚNeoMMEDenseHeadNrÎ   r$  Ú	dense_dimr7   c                 óò  — |                      d¦  «                             |j        ¦  «                             |j        ¦  «        }||z                       d¦  «        |                     d¦  «                             d¦  «        z  }|€t          j        |d¬¦  «        S d|cxk     r|j        d         k    s"n t          d|j        d         › d|› �¦  «        ‚t          j        |dd |…f         d¬¦  «        S )	Nro   r1   g•Ö&è.>r†   r   zdense_dim must be in 1..z (the pooled width), got .)
r�  rÜ   rq   rä   r…   rÿ   Ú	clamp_minrý   rþ   rh   )r_   rÎ   r$  rÊ  Úexpanded_maskÚpooleds         rH   rÀ   zNeoMMEDenseHead.forwardÙ  sþ   € ð '×0Ò0°Ñ4Ô4×;Ò;¸MÔ<OÑPÔP×SÒSÐTaÔTgÑhÔhˆØ -Ñ/×4Ò4°QÑ7Ô7¸-×:KÒ:KÈAÑ:NÔ:N×:XÒ:XÐY]Ñ:^Ô:^Ñ^ˆØÐÝ”;˜v¨2Ð.Ñ.Ô.Ð.à�9Ð0Ð0Ò0Ð0 ¤¨RÔ 0Ò0Ð0Ð0Ð0ÝÐn¸¼ÀRÔ8HÐnÐnÐclÐnÐnÑoÔoÐoÝŒ{˜6 # z¨	 z /Ô2¸Ð;Ñ;Ô;Ð;rJ   rŸ   )rN   rO   rP   rŒ   rÂ   rR   rÀ   rT   rJ   rH   rÉ  rÉ  Ø  s\   € € € € € àaeð
<ð 
<Ø"œ\ð
<Ø;@¼<ð
<ØTWÐZ^ÑT^ð
<à	Œð
<ð 
<ð 
<ð 
<ð 
<ð 
<rJ   rÉ  z½
    The NeoMME model with multi-vector and dense retrieval heads. One forward pass can return token embeddings for
    MaxSim scoring and mean-pooled embeddings for cosine similarity.
    c                   óÒ   ‡ — e Zd Zdefˆ fd„Zee	 	 	 	 	 	 	 ddej        dz  dej	        dz  dej        dz  dej	        dz  d	e
d
e
dedz  dee         defd„¦   «         ¦   «         Zˆ xZS )ÚNeoMMEForRetrievalr±   c                 óè   •— t          ¦   «                              |¦  «         t          |¦  «        | _        t	          |¦  «        | _        t          ¦   «         | _        |                      ¦   «          d S rŸ   )	r]   r^   rd  r³  rÃ  Úmulti_vector_headrÉ  Ú
dense_headrŽ  r½   s     €rH   r^   zNeoMMEForRetrieval.__init__í  s]   ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ý  Ñ(Ô(ˆŒ
Ý!6°vÑ!>Ô!>ˆÔÝ)Ñ+Ô+ˆŒØ�ŠÑÔÐÐÐrJ   NTr¾   r$  rÒ   rY   Úoutput_multivectorÚoutput_denserÊ  r[   r7   c                 ój  — |s|st          d¦  «        ‚ | j        d||||dœ|¤Ž}	|	j        }
|€3t          j        |
j        dd…         t          j        |
j        ¬¦  «        }|r|                      |
|¦  «        nd}|r|  	                    |
||¦  «        nd}t          |||
|	j        |	j        ¬¦  «        S )aà  
        output_multivector (`bool`, *optional*, defaults to `True`):
            Whether to return token embeddings for late-interaction retrieval.
        output_dense (`bool`, *optional*, defaults to `True`):
            Whether to return one mean-pooled dense embedding per input.
        dense_dim (`int`, *optional*):
            Width of the Matryoshka prefix to return for dense embeddings. The model truncates the pooled vector
            before normalizing it.
        zCAt least one of `output_multivector` or `output_dense` must be Truer·  Nr'   )r…   rÞ   )r¿  rÀ  r’  rÎ   rN  rT   )rh   r³  r’  rŒ   Úonesrq   r¦   rÞ   rÒ  rÓ  r¾  rÎ   rN  )r_   r¾   r$  rÒ   rY   rÔ  rÕ  rÊ  r[   r»  rÎ   r¿  rÀ  s                rH   rÀ   zNeoMMEForRetrieval.forwardô  sü   € ð, #ð 	d lð 	dÝÐbÑcÔcÐcà#- 4¤:ð $
ØØ)Ø%Ø%ð	$
ð $
ð
 ð$
ð $
ˆð  Ô1ˆØÐ!Ý"œZ¨Ô(;¸B¸Q¸BÔ(?ÅuÄzÐZgÔZnÐoÑoÔoˆNàN`Ðj�T×+Ò+¨M¸>ÑJÔJÐJÐfjˆ
ØXdÐn˜4Ÿ?š?¨=¸.È)ÑTÔTÐTÐjnÐÝ'Ø!Ø-Ø+Ø!Ô/ØÔ)ð
ñ 
ô 
ð 	
rJ   )NNNNTTN)rN   rO   rP   r2   r^   r#   r   rŒ   rÁ   rÂ   r¦   rR   r   r   r¾  rÀ   r©   rª   s   @rH   rÐ  rÐ  æ  s  ø€ € € € € ð˜|ð ð ð ð ð ð ð Øð .2Ø.2Ø04Ø,0Ø#'Ø!Ø $ð*
ð *
àÔ# dÑ*ð*
ð œ tÑ+ð*
ð Ô&¨Ñ-ð	*
ð
 ”l TÑ)ð*
ð !ð*
ð ð*
ð ˜‘:ð*
ð Ð+Ô,ð*
ð 
"ð*
ð *
ð *
ñ „^ñ Ôð*
ð *
ð *
ð *
ð *
rJ   rÐ  )rf  rÐ  rW   rd  rL  )fr?   Úcollections.abcr   Údataclassesr   Útypingr   rŒ   Útorch.nn.functionalr   r   rý   Útorchvision.transforms.v2rv   Ú r	   r_  Úimage_processing_backendsr
   Úimage_processing_utilsr   Úimage_transformsr   r   Úimage_utilsr   r   r   Úmasking_utilsr   r   Úmodeling_layersr   Úmodeling_outputsr   r   r   Úmodeling_rope_utilsr   Úmodeling_utilsr   r   Úprocessing_utilsr   r   Úutilsr   r   r   r   r    Úutils.constantsr!   r"   Úutils.genericr#   r$   Úutils.output_capturingr%   Úutils.type_validatorsr&   Úgemma4.modeling_gemma4r(   Úgpt_neox.modeling_gpt_neoxr)   Úlaguna.modeling_lagunar*   Ú lfm2_vl.image_processing_lfm2_vlr+   Úllama.modeling_llamar,   r-   Ú"muse_glimmer.modeling_muse_glimmerr.   r/   Únemotron.modeling_nemotronr0   Úconfiguration_neommer2   rR   r¤   rI   rL   rW   r¬   rx  r°   rÄ   rÐ   rñ   r  r  r:  r<  rL  rd  rf  r¾  rÃ  rÉ  rÐ  Ú__all__rT   rJ   rH   ú<module>rö     s:  ðð €€€Ø $Ð $Ð $Ð $Ð $Ð $Ø !Ð !Ð !Ð !Ð !Ð !Ø Ð Ð Ð Ð Ð à €€€Ø Ð Ð Ð Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7à &Ð &Ð &Ð &Ð &Ð &Ø ;Ð ;Ð ;Ð ;Ð ;Ð ;Ø 2Ð 2Ð 2Ð 2Ð 2Ð 2Ø EÐ EÐ EÐ EÐ EÐ EÐ EÐ EØ CÐ CÐ CÐ CÐ CÐ CÐ CÐ CÐ CÐ CØ `Ð `Ð `Ð `Ð `Ð `Ð `Ð `Ø 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø [Ð [Ð [Ð [Ð [Ð [Ð [Ð [Ð [Ð [Ø 6Ð 6Ð 6Ð 6Ð 6Ð 6Ø FÐ FÐ FÐ FÐ FÐ FÐ FÐ FØ 4Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4ðð ð ð ð ð ð ð ð ð ð ð ð ð ð MÐ LÐ LÐ LÐ LÐ LÐ LÐ LØ =Ð =Ð =Ð =Ð =Ð =Ð =Ð =Ø 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø 1Ð 1Ð 1Ð 1Ð 1Ð 1Ø 2Ð 2Ð 2Ð 2Ð 2Ð 2Ø =Ð =Ð =Ð =Ð =Ð =Ø :Ð :Ð :Ð :Ð :Ð :Ø GÐ GÐ GÐ GÐ GÐ GØ EÐ EÐ EÐ EÐ EÐ EÐ EÐ EØ eÐ eÐ eÐ eÐ eÐ eÐ eÐ eØ 4Ð 4Ð 4Ð 4Ð 4Ð 4Ø .Ð .Ð .Ð .Ð .Ð .ð&) 3ð &)¨sð &)¸cÀD¹jð &)ÐPXÐ[_ÑP_ð &)ÐdiÐjmÐorÐjrÔdsð &)ð &)ð &)ð &)ðR	4ð 	4ð 	4ð 	4ð 	4 °Uð 	4ñ 	4ô 	4ð 	4ð ðJDð JDð JDð JDð JDÐ-ñ JDô JDñ „ðJDðZ	ð 	ð 	ð 	ð 	�Mñ 	ô 	ð 	ð	Jð 	Jð 	Jð 	Jð 	J�r”yñ 	Jô 	Jð 	Jð:ð :ð :ð :ð :˜BœIñ :ô :ð :ð3ð 3ð 3ð 3ð 3Ð1ñ 3ô 3ð 3ðD?ð ?ð ?ð ?ð ? 2¤9ñ ?ô ?ð ?ðWð Wð Wð Wð W 2¤9ñ Wô Wð Wð ðF)ð F)ð F)ð F)ð F)Ð.ñ F)ô F)ñ ÔðF)ðR	ð 	ð 	ð 	ð 	�ñ 	ô 	ð 	ð#Eð #Eð #Eð #Eð #EÐ3ñ #Eô #Eð #EðL8ð 8ð 8ð 8ð 8Ð6ñ 8ô 8ð 8ðv €ððñ ô ð
p@ð p@ð p@ð p@ð p@Ð'ñ p@ô p@ñô ð
p@ðf €ððñ ô ð
1
ð 1
ð 1
ð 1
ð 1
Ð-ñ 1
ô 1
ñô ð
1
ðh €ððñ ô ð
 ð6ð 6ð 6ð 6ð 6˜ñ 6ô 6ñ „ñô ð6ð 	Qð 	Qð 	Qð 	Qð 	Q˜BœIñ 	Qô 	Qð 	Qð<ð <ð <ð <ð <�b”iñ <ô <ð <ð €ððñ ô ð4
ð 4
ð 4
ð 4
ð 4
Ð.ñ 4
ô 4
ñô ð4
ðnð ð €€€rJ   