tjKdZddlmZddlmZddlZddlmZddlmZddl m Z dd l m Z dd l mZdd lmZmZdd lmZdd lmZmZmZddlmZmZmZddlmZddlmZej e!Z"dZ#Gddej$Z%dZ&d/dZ' d0dej$dej(dej(dej(dej(dzde)de)fdZ*Gd d!ej$Z+Gd"d#ej$Z,Gd$d%ej$Z-Gd&d'e Z.Gd(d)ej$Z/eGd*d+eZ0d,Z1eGd-d.e0Z2d.d+gZ3dS)1zPyTorch Pixtral model.)Callable)OptionalN)nn)ACT2FN)GradientCheckpointingLayer)BaseModelOutput)dynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringlogging)is_flash_attention_requestedmaybe_autocastmerge_with_config_defaults)capture_outputs)PixtralVisionConfigcg}|D]}|jdd\}}tjtj|tj|d}tj|ddddd\}}||z|z} || dddftj|S)Nij)indexingdimr) shapetorchmeshgridarangestackreshapechunkappendcat) patch_embeds_list max_width positionspatchheightwidthmeshh_gridv_grididss j/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/pixtral/modeling_pixtral.pyposition_ids_in_meshgridr3%sI"$$ BCC( ~el622EL4G4GRVWWWTr222::2qAAGG2NNy 6)QQQT#### 9Y  c eZdZUdZejed<d deffd Ze ddedzde dde dzd e d e ffd Zejed ZxZS)PixtralRotaryEmbeddinga The key with pixtral embedding is just that you have a frequency for each pixel positions. If you have height x width pixels (or embedding pixels), then the frequency used for ROPE is given by indexing the pre_computed frequency on the width and height. What you output is of dimension (batch, height * width, dim) with dim the embed dim. This simply means that for each image hidden state, you are going to add a corresponding positional embedding, based on its index in the grid. inv_freqNconfigct||_|jjd|_|j}|jdkr$t |jjd|jd||j|\}}| d|d| d| ddS) N rope_typedefaultz7 does not support non-default RoPE, but got `rope_type=`r7F) persistentoriginal_inv_freq) super__init__r8rope_parametersr:compute_default_rope_parameters ValueError __class____name__register_bufferclone)selfr8device layer_type rope_init_fnr7attention_scalingrDs r2r@zPixtralRotaryEmbedding.__init__>s  4[A!%!E >Y & &>*ttcgcqttt '3l4;&G&G## ZeDDD 0(..2B2BuUUUUUr4rIz torch.deviceseq_lenreturnz torch.Tensorc|jd}t|ddp|j|jz}d}|j|jz}t j|}t j|}d|t jd|d|z zz } t j || ddd} t j || ddd} t j | dddddf d|d| dddddf |ddgd  d|dz} t j | | fd } | |fS) a Computes the inverse frequencies according to the original RoPE implementation Args: config ([`~transformers.PreTrainedConfig`]): The model configuration. device (`torch.device`): The device to use for initialization of the inverse frequencies. seq_len (`int`, *optional*): The current sequence length. Unused for this type of RoPE. Returns: Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE). rope_thetahead_dimNg?rrrrr) rAgetattr hidden_sizenum_attention_heads image_size patch_sizer r"floatouterr'repeatr$) r8rIrMbaserattention_factormax_patches_per_sidehwfreqsfreqs_hfreqs_wr7s r2rBz6PixtralRotaryEmbedding.compute_default_rope_parametersNs&%l3fj$//c63EIc3c &0F4EE L- . . L- . .t QQ 7 7 = = ? ?# EFG+asss,,2244+aqt!t--335594 #**1.BAFFaaa #**+?AFF     '"cQh   9h1r:::)))r4c|j|}t|jjtr|jjdkr |jjnd}t |d5|}|}|}dddn #1swxYwY||j ||j fS)NmpscpuF) device_typeenableddtype) r7 isinstancerItypestrrcossintorh)rHx position_idsr_reembrlrms r2forwardzPixtralRotaryEmbedding.forward{s l+'1!(-'E'Ek!(-[`J`J`ahmmfk  U C C C  C''))C''))C                vvAGv$$cff17f&;&;;;s+BBBNN)NNN)rE __module__ __qualname____doc__r Tensor__annotations__rr@ staticmethodrinttuplerWrBno_gradr rr __classcell__rDs@r2r6r60s  lVV2VVVVVV -1+/"****#d***(**t** ~u$ % ******\**XU]__<<_<<<<