Zj9SSKrSSKJr SSKJr SSKJrJr SSKJr SSK J r SSK J r SS K Jr SS KJrJrJrJrJr S S KJrJr S S KJr S SKJr S SKJrJrJrJ r J!r! \RD"\#5r$\"SS9\"SS\555r%"SS\ 5r&"SS\!5r'\"SS9\"SS\555r("SS\5r)"SS\5r*"SS\5r+"S S!\5r,"S"S#\5r-/S$Qr.g)%N)strict)nn)Cache DynamicCache)GenerationConfig)FlashAttentionKwargs)BaseModelOutputWithPooling)Unpack)TransformersKwargsauto_docstringcan_return_tupleloggingtorch_compilable_check)Idefics3ConfigIdefics3VisionConfig)Idefics3ImageProcessor)Idefics3ImageProcessorPil)Idefics3BaseModelOutputWithPast Idefics3ForConditionalGeneration Idefics3ModelIdefics3PreTrainedModelIdefics3VisionTransformerz$HuggingFaceTB/SmolVLM2-2.2B-Instruct) checkpointc\rSrSrSrSrSrg)SmolVLMVisionConfig*aS Example: ```python >>> from transformers.models.smolvlm.modeling_smolvlm import SmolVLMVisionTransformer >>> from transformers.models.smolvlm.configuration_smolvlm import SmolVLMVisionConfig >>> # Initializing a SmolVLMVisionConfig with google/siglip-so400m-patch14-384 style configuration >>> configuration = SmolVLMVisionConfig() >>> # Initializing a SmolVLMVisionTransformer (with random weights) from the google/siglip-so400m-patch14-384 style configuration >>> model = SmolVLMVisionTransformer(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```smolvlm_visionN__name__ __module__ __qualname____firstlineno____doc__ model_type__static_attributes__r |/root/GenerationalWealth/GenerationalWealth/venv/lib/python3.13/site-packages/transformers/models/smolvlm/modular_smolvlm.pyrr*s ""Jr)rc\rSrSrSrg)SmolVLMPreTrainedModelAr Nr"r#r$r%r(r r)r*r,r,Ar)r,c\rSrSrSrg)SmolVLMVisionTransformerEr Nr.r r)r*r1r1Er/r)r1c\rSrSrSrSrSrg) SmolVLMConfigIa scale_factor (`int`, *optional*, defaults to 2): The scale factor for the image encoder. Example: ```python >>> from transformers import SmolVLMModel, SmolVLMConfig >>> # Initializing configuration >>> configuration = SmolVLMConfig() >>> # Initializing a model from the configuration >>> model = SmolVLMModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```smolvlmr Nr!r r)r*r4r4Is  Jr)r4c\rSrSrSrg)SmolVLMImageProcessor^r Nr.r r)r*r8r8^r/r)r8c\rSrSrSrg)SmolVLMImageProcessorPilbr Nr.r r)r*r;r;br/r)r;c\rSrSrSrg)SmolVLMBaseModelOutputWithPastfr Nr.r r)r*r>r>fr/r)r>c \rSrSrSrS\R S\RS\R4Sjr\ \ "SS9SS \RS \R S -S \ \ S \\-4Sjj55r\ \ "SS9\ SS\R S -S\RS -S\R S -S\S -S\RS -S \RS -S \R$S -S\RS -S\S -S \ \S \\-4Sjj555rSrg ) SmolVLMModeljz A subclass of Idefics3Model. We do *not* remove or block the call to inputs_merger in forward. Instead, we override inputs_merger here with custom logic. input_ids inputs_embedsimage_hidden_statesc URupEnUc^X R5"[R"URR [R URS95:HnUSnOXRR :HnURSS9n[[R"Xu-S:H5S5 Xu-n[RRRURSS9SSS9n U SS n URS S9n U S- U-n U S- U-n U RS5U -n[R "U5nX>UXSS24X'[R""URS 5X5nU$) Ndtypedevice).rdimrzCAt least one sample has tokens not divisible by patch_size.)rJr)value)shapeget_input_embeddingstorchtensorconfigimage_token_idlongrIsumrallr functionalpadcumsum unsqueeze zeros_likewhere)selfrCrDrE_ patch_size image_masknum_image_tokensblocks_per_sampleoffsets block_offsetrow_cum chunk_idx local_idx block_idx image_embeds merged_embedss r* inputs_mergerSmolVLMModel.inputs_mergerps}/44q  &*C*C*E T[[77uzzR_RfRfg+J$F+J"kk&@&@@J%>>a>0 II&3q8 9 Q -:((%%))*;*B*Bq*B*I6YZ)[s| ###+q[Z/ q[J.  **1- 9 '' 6 #67LiNcef7f#g   J$8$8$P>P; e#TZZ8 #(()@Z?Y((+Y?jj#**$ $8#<# (batch_size, max_num_images, 3, max_heights, max_widths) where max_num_images is the maximum number of images among the batch_size samples in the batch. Padding images are not needed beyond padding the pixel_values at the entrance of the model. For efficiency, we only pass through the vision_model's forward the real images by discarding the padding images i.e. pixel_values of size (image_batch_size, 3, height, width) where image_batch_size would be 7 when num_images_per_sample=[1, 3, 1, 2] and max_num_images would be 3. attention_mask position_idspast_key_values use_cachec LUR(a9URR(aU (a[R S5 Sn UbUR upOUbUR upn O [ S5eU (aUc[URS9nUc9URR5"U5RUR5nUbUb [ S5eUb6URXgSS9RnURUR5nO'Ub$URURURS9nUbURUUUS 9nUR"S UUUUU S .U D6n[!UR"UR$UR&UR(US 9$) NzZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Fz5You have to specify either input_ids or inputs_embeds)rSzMYou cannot specify both pixel_values and image_hidden_states at the same timeT)rzrG)rCrDrE)rDrrrr)rr hidden_states attentionsrEr )training text_modelgradient_checkpointinglogger warning_oncerO ValueErrorrrSrPr{rIrrrHrlr>rrrr)r^rCrrrrDrorprErrqr seq_lengthr_outputss r*forwardSmolVLMModel.forwards4 ==T__CC   l I  %.__ "J  &(5(;(; %JATU U 0*$++>O   OO@@B9MPPQZQaQabM  #(;(Glm m  #"&"9"9#:#m #6"8"89M9M"N  ,"5"8"8tzzR_RfRf"8"g   * ..#+$7/M // ')%+    .%77#33!//)) 3   r)r )N) NNNNNNNNN)r"r#r$r%r&rQ LongTensorTensorrlrr FloatTensorr r tupler rr BoolTensorrr r>rr(r r)r*rArAjs )):?,,]b]i]iBm 9=2''2$..52+, 2 + + 22h  .2.204(,26158<8[TU]U5 [U5Ul[R "U5URR l[R"URRURRSS9Ul UR5 g)NF)bias)super__init__rAmodelrfrom_model_configrgeneration_configrLinear text_config hidden_size vocab_sizelm_head post_init)r^rS __class__s r*r(SmolVLMForConditionalGeneration.__init__sq  !&) 2B2T2TU[2\ /yy!3!3!?!?ASASA^A^ejk  r)c &>[TU]"S0UD6 g)a pixel_attention_mask (`torch.Tensor` of shape `(batch_size, image_size, image_size)`, *optional*): Mask to avoid performing attention on padding pixel indices. image_hidden_states (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)`): The hidden states of the image encoder after modality projection. 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 `model.image_token_id`. Tokens with indices set to `model.image_token_id` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. Example: ```python >>> import httpx >>> from io import BytesIO >>> import torch >>> from PIL import Image >>> from io import BytesIO >>> from transformers import AutoProcessor, AutoModelForImageTextToText >>> from transformers.image_utils import load_image >>> # Note that passing the image urls (instead of the actual pil images) to the processor is also possible >>> image1 = load_image("https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg") >>> image2 = load_image("https://cdn.britannica.com/59/94459-050-DBA42467/Skyline-Chicago.jpg") >>> image3 = load_image("https://cdn.britannica.com/68/170868-050-8DDE8263/Golden-Gate-Bridge-San-Francisco.jpg") >>> processor = AutoProcessor.from_pretrained("HuggingFaceTB/SmolVLM2-2.2B-Instruct") >>> model = AutoModelForImageTextToText.from_pretrained("HuggingFaceTB/SmolVLM2-2.2B-Instruct", dtype=torch.bfloat16, device_map="auto") >>> # Create inputs >>> messages = [ ... { ... "role": "user", ... "content": [ ... {"type": "video", "path": path/to/video}, ... {"type": "text", "text": "What is happening in this video?"}, ... ] ... } ... ] >>> inputs = processor.apply_chat_template([messages], add_generation_prompt=True) >>> # Generate >>> generated_ids = model.generate(**inputs, max_new_tokens=256) >>> generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True) >>> print(generated_texts) ```Nr )rr)r^ super_kwargsrs r*r'SmolVLMForConditionalGeneration.forward$sd ','r))rr) r"r#r$r%_tied_weights_keysrrr( __classcell__)rs@r*rrs*,RS2(2(r)r)rr4r8r;rr,rAr1)/rQhuggingface_hub.dataclassesrr cache_utilsrr generationrmodeling_flash_attention_utilsr modeling_outputsr processing_utilsr utilsr r rrridefics3.configuration_idefics3rr"idefics3.image_processing_idefics3r&idefics3.image_processing_pil_idefics3ridefics3.modeling_idefics3rrrrr get_loggerr"rrr,r1r4r8r;r>rAr__all__r r)r*rs" ..*B:&jjRGN   H %AB"."C"* 4  8 ABNC& 2  8  %D m =m `<(&F<(~ r)