o Zh<@sdZddlmZmZmZddlmZddlmZm Z m Z ddl m Z m Z mZmZmZmZddlmZmZmZddlmZeeZd Zd d ed Dd d edDZGdddeZGddde ZGddde ddZ de!fddZ"ddZ#ddZ$ddZ%Gd d!d!eZ&d!gZ'd"S)#z Processor class for PaliGemma. )ListOptionalUnion) BatchFeature) ImageInputis_valid_imagemake_flat_list_of_images) ImagesKwargsProcessingKwargsProcessorMixin TextKwargsUnpack!_validate_images_text_input_order) AddedTokenPreTokenizedInput TextInput)loggingzcCg|] }d|ddqS)z4>.0irra/var/www/auris/lib/python3.10/site-packages/transformers/models/paligemma/processing_paligemma.py *ricCr)z3rrrrrrr*rc@s.eZdZUeeeeeeeefed<dS)PaliGemmaTextKwargssuffixN) __name__ __module__ __qualname__rrrrr__annotations__rrrrr-s $rc@seZdZUeeed<dS)PaliGemmaImagesKwargsZdo_convert_rgbN)r r!r"rboolr#rrrrr$1s r$c@s0eZdZUeed<eed<ddiddidZdS) PaliGemmaProcessorKwargs text_kwargs images_kwargspaddingFZ data_formatZchannels_first)r'r(N)r r!r"rr#r$ _defaultsrrrrr&5s  r&F)totalreturncCst|to |dS)Nhttp) isinstancestr startswith)valrrris_urlCsr2cCst|pt|SN)r2relemrrris_image_or_image_urlHsr6cCst|tpt|Sr3)r.r/r6r4rrr_is_str_or_imageLsr7cCs|||||dS)aZ Builds a string from the input prompt and image tokens. For example, for the call: build_string_from_input( prompt="Prefix str" bos_token="", image_seq_len=3, image_token="", ) The output will be: "Initial str" Args: prompt (`List[Union[str, ImageInput]]`): The input prompt. bos_token (`str`): The beginning of sentence token. image_seq_len (`int`): The length of the image sequence. image_token (`str`): The image token. num_images (`int`): Number of images in the prompt.  rprompt bos_tokenZ image_seq_len image_tokenZ num_imagesrrrbuild_string_from_inputPsr=c seZdZdZddgZdgZdZdZ   dfdd Z    dd e d e e e e e e e fd eed efddZddZddZeddZZS)PaliGemmaProcessora Constructs a PaliGemma processor which wraps a PaliGemma image processor and a PaliGemma tokenizer into a single processor. [`PaliGemmaProcessor`] offers all the functionalities of [`SiglipImageProcessor`] and [`GemmaTokenizerFast`]. See the [`~PaliGemmaProcessor.__call__`] and [`~PaliGemmaProcessor.decode`] for more information. Args: image_processor ([`SiglipImageProcessor`], *optional*): The image processor is a required input. tokenizer ([`GemmaTokenizerFast`], *optional*): The tokenizer is a required input. chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages in a chat into a tokenizable string. image_processor tokenizer chat_template)ZSiglipImageProcessorZSiglipImageProcessorFast)ZGemmaTokenizerZGemmaTokenizerFastNc s|durtd|durtdt|dstd|j|_t|ds=ttddd}d |gi}|||t|_t|_n|j|_|j|_| t d|_ d|_ t j|||d dS) Nz)You need to specify an `image_processor`.z"You need to specify a `tokenizer`.image_seq_lengthz;Image processor is missing an `image_seq_length` attribute.r<FT) normalizedZspecialZadditional_special_tokens)rA) ValueErrorhasattrrBr IMAGE_TOKENZadd_special_tokensZconvert_tokens_to_idsZimage_token_idr<Z add_tokens EXTRA_TOKENSZ add_bos_tokenZ add_eos_tokensuper__init__)selfr?r@rAkwargsr<Z tokens_to_add __class__rrrI{s&      zPaliGemmaProcessor.__init__imagestextrKr,c st||\}}jtfdjji|}|ddd}|dur"dnd}|dur,td|dur7tdd }t |r?|g}n t |t rKt |d rK |dur |dur t d d |Dst d t |trt |trt|t|krtdt|dt|dt|r|gg}n1t |t tfrt|d rdd|D}nt |t tfrt |d t tfrt|d d stdfddt||D} t|}n=g} |D]1} | ttj} | t} | dkr| ttnd }| d|jj| |d} | | qdd| D} |durt |r|g}|dur'fdd|D}j|fi|dd}|ddd}j| f||d|d}j| |dgdi|d|i}|rq|d|dd kd }|d!|it||d"S)#ah Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text` and `kwargs` arguments to GemmaTokenizerFast's [`~GemmaTokenizerFast.__call__`] if `text` is not `None` to encode the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to SiglipImageProcessor's [`~SiglipImageProcessor.__call__`] if `images` is not `None`. Please refer to the docstring of the above two methods for more information. The usage for PaliGemma fine-tuning preparation is slightly different than usual. suffix passed are suffixes to the prompt in `text`, and will be placed after the prompt. This is because attention is handled differently for the prefix and the suffix. For instance, ```python image = PIL_cow_image prompt = "answer en Where is the cow standing?" suffix = "on the beach" inputs = processor(text=prompt, images=image, suffix=suffix) ``` Here `inputs` will contain the `input_ids` and `token_type_ids` that follow ```python inputs["input_ids"][:, 256:] # tensor([[ 2, 6006, 603, 573, 13910, 9980, 235336, 108, 477, 573, 8318]]) inputs["token_type_ids"][:, 256:] tensor([[0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1]]) ``` Meaning the last three tokens are of "label" ("suffix") type while the other ones are of "prefix" type. Args: images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`): The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch tensor. In case of a NumPy array/PyTorch tensor, each image should be of shape (C, H, W), where C is a number of channels, H and W are image height and width. text (`str`, `List[str]`, `List[List[str]]`): The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set `is_split_into_words=True` (to lift the ambiguity with a batch of sequences). return_tensors (`str` or [`~utils.TensorType`], *optional*): If set, will return tensors of a particular framework. Acceptable values are: - `'tf'`: Return TensorFlow `tf.constant` objects. - `'pt'`: Return PyTorch `torch.Tensor` objects. - `'np'`: Return NumPy `np.ndarray` objects. - `'jax'`: Return JAX `jnp.ndarray` objects. suffix (`str`, `List[str]`, `List[List[str]]`): The suffixes or batch of suffixes to be encoded. Only necessary for finetuning. See https://github.com/google-research/big_vision/blob/main/big_vision/configs/proj/paligemma/README.md for more information. If your prompt is " What is on the image", the suffix corresponds to the expected prediction "a cow sitting on a bench". Returns: [`BatchFeature`]: A [`BatchFeature`] with the following fields: - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`. If `suffix` is provided, the `input_ids` will also contain the suffix input ids. - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not `None`). - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`. - **labels** -- Labels compatible with training if `suffix` is not None Ztokenizer_init_kwargsr'rNTFzF`images` are expected as arguments to a `PaliGemmaProcessor` instance.z]You are using PaliGemma without a text prefix. It will perform as a picture-captioning model.rcss|]}t|vVqdSr3)rFrsamplerrr sz.PaliGemmaProcessor.__call__..aLYou are passing both `text` and `images` to `PaliGemmaProcessor`. The processor expects special image tokens in the text, as many tokens as there are images per each text. It is recommended to add `` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens.z Received z images for zK prompts. Each prompt should be associated with an image or list of images.cSsg|]}|gqSrr)rimagerrrrsz/PaliGemmaProcessor.__call__..zAimages must be an image, list of images or list of list of imagesc s:g|]\}}t|jjjtt|trt|nddqS)r9)r=r@r;rBrFr.listlen)rr:Z image_listrJrrrscSsg|]}|dqS)r8rrQrrrr%scsg|]}|jjqSr)r@Z eos_token)rZsfxrXrrr*sr( pixel_valuesreturn_tensors)Z text_pairreturn_token_type_idsrT)Z modalitiesZ input_idsZtoken_type_idsilabels)dataZ tensor_type)rZ _merge_kwargsr&r@Z init_kwargspoprDloggerZ warning_oncer7r.rVanywarningrrWrtuplezipr replacerFrBrfindr;appendr?Z_check_special_mm_tokensZ masked_fillupdater)rJrNrOZaudioZvideosrKZ output_kwargsrr\Z input_stringsZexpanded_samplesrRZexpanded_sampleZbos_rfind_indexZ bos_indexrZr[ZinputsZ return_datar]rrXr__call__sB        zPaliGemmaProcessor.__call__cO|jj|i|S)z This method forwards all its arguments to GemmaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please refer to the docstring of this method for more information. )r@ batch_decoderJargsrKrrrrk>zPaliGemmaProcessor.batch_decodecOrj)z This method forwards all its arguments to GemmaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to the docstring of this method for more information. )r@decoderlrrrroFrnzPaliGemmaProcessor.decodecCs"|jj}|jj}tt||Sr3)r@model_input_namesr?rVdictfromkeys)rJZtokenizer_input_namesZimage_processor_input_namesrrrrpMsz$PaliGemmaProcessor.model_input_names)NNN)NNNN)r r!r"__doc__ attributesZ valid_kwargsZimage_processor_classZtokenizer_classrIrrrrrrr&rrirkropropertyrp __classcell__rrrLrr>fs8" $r>N)(rstypingrrrZfeature_extraction_utilsrZ image_utilsrrr Zprocessing_utilsr r r r rrZtokenization_utils_baserrrutilsrZ get_loggerr r`rFrangerGrr$r&r%r2r6r7r=r>__all__rrrrs(    $ p