$//r c & \ rS rSr% \\ \S' Srg)Idefics2ImagesKwargs1
image_seq_len N)__name__
__module__ __qualname____firstlineno__r int__annotations____static_attributes__r& r r r# r# 1 s C= r r# F)totalc 2 \ rS rSr% \\S' SSSS.0 S.rSrg) Idefics2ProcessorKwargs5
images_kwargsTF)add_special_tokenspaddingis_split_into_words) text_kwargsr2 r& N)r' r( r) r* r# r, _defaultsr- r& r r r0 r0 5 s$ '' #' #(
Ir r0 c
^ \ rS rSrSrSS/rSrSr SS\S\ \
4U 4S jjjr S
r SS \
\\\ \\\ 4 S \
\S
\\ \S
4 S\\ S\4S jjrS rS r\S 5 rSrU =r$ )Idefics2ProcessorB a
Constructs a IDEFICS2 processor which wraps a LLama tokenizer and IDEFICS2 image processor into a single processor.
[`IdeficsProcessor`] offers all the functionalities of [`Idefics2ImageProcessor`] and [`LlamaTokenizerFast`]. See
the docstring of [`~IdeficsProcessor.__call__`] and [`~IdeficsProcessor.decode`] for more information.
Args:
image_processor (`Idefics2ImageProcessor`):
An instance of [`Idefics2ImageProcessor`]. The image processor is a required input.
tokenizer (`PreTrainedTokenizerBase`, *optional*):
An instance of [`PreTrainedTokenizerBase`]. This should correspond with the model's text model. The tokenizer is a required input.
image_seq_len (`int`, *optional*, defaults to 64):
The length of the image sequence i.e. the number of tokens per image in the input.
This parameter is used to build the string from the input prompt and image tokens and should match the
config.perceiver_config.resampler_n_latents value for the model used.
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 tokenizerIdefics2ImageProcessor
AutoTokenizerr%
chat_templatec J > Uc [ S5 eUc [ S5 e[ US5 ( d [ SSSS9R U l [ SSSS9R U l S U R U R
/0nUR
U5 UR U R
5 U l O3UR U l UR
U l UR U l [ S
SSS9U l
UR
S U R /05 X0l [ TU ]5 XUS 9 g ) Nz)You need to specify an `image_processor`.z"You need to specify a `tokenizer`. image_tokenzFT)
normalizedspecialzadditional_special_tokensz)r? )
ValueErrorhasattrr contentfake_image_tokenrA r3 convert_tokens_to_idsimage_token_idimage_boundary_tokenend_of_utterance_tokenr% super__init__)selfr; r< r% r? kwargs
tokens_to_add __class__s r rN Idefics2Processor.__init__Z s " HI I AB By-00$./JW\fj$k$s$sD ! ))t T \ \D 84;P;PRVRbRb:c dM ( ( 7"+"A"A$BRBR"SD $-$B$BD ! ( 4 4D "+":":D &01ERWae&f#$$&ADD_D_C`%ab *
=Qr c / nU Hn n/ nU HR n[ U5 ( a UR U5 M& [ U5 ( d M8 UR [ U5 5 MT UR U5 Mp U$ r )r
appendr r )rO prompts
prompt_imagespromptimagesr s r _extract_images_from_prompts.Idefics2Processor._extract_images_from_promptss si
FF !$''MM$'D\\MM*T"23
( r rY textr rP r c Uc Uc [ S5 eU R " [ 4SU R R 0UD6nUS R SS5 nUb UO U R nUS R SS5 n/ n 0 n
UGb5 [ U[ 5 ( a U/nO8[ U[ 5 ( d# [ US [ 5 ( d [ S 5 eU R n U R n U X- U 3n
U R R ( a
U
S
- n
US
- n/ nU H\ nU R UR U 5 5 UR! X5 nUR! U U 3U 5 nUR U5 M^ U R " U40 US D6nU R# UUS /S 9 U
R% U5 UGb [' U5 ( a U//nGO[ U[ [( 45 ( a [' US 5 ( a Ub [+ U 5 [- U5 :w a* [ S
W S[+ U 5 SU S[- U5 S3 5 eS/[ [/ U 5 5 - n[1 [- U 5 5 Vs/ sH nUUU UUS- PM nnO^U/nOZ[ U[ [( 45 ( d? [ US [ [( 45 ( d! [' US S 5 ( d [ S5 eU Vs/ sH n[- U5 PM nnUb UU :X d [ SU SU S35 eU VVs/ sH o Vs/ sH n[3 U5 PM snPM nnnU R " U40 US D6nU
R% U5 [5 XS9$ s snf s snf s snf s snnf )a
Processes the input prompts and returns a BatchEncoding.
Example:
```python
>>> import requests
>>> from transformers import Idefics2Processor
>>> from transformers.image_utils import load_image
>>> processor = Idefics2Processor.from_pretrained("HuggingFaceM4/idefics2-8b", image_seq_len=2)
>>> processor.image_processor.do_image_splitting = False # Force as False to simplify the example
>>> url1 = "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
>>> url2 = "https://cdn.britannica.com/59/94459-050-DBA42467/Skyline-Chicago.jpg"
>>> image1, image2 = load_image(url1), load_image(url2)
>>> images = [[image1], [image2]]
>>> text = [
... "In this image, we see",
... "bla bla bla",
... ]
>>> outputs = processor(images=images, text=text, return_tensors="pt", padding=True)
>>> input_ids = outputs.input_ids
>>> input_tokens = processor.tokenizer.batch_decode(input_ids)
>>> print(input_tokens)
[' In this image, we see', ' bla bla bla']
```
Args:
images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `list[PIL.Image.Image]`, `list[np.ndarray]`, `list[torch.Tensor]`, *optional*):
The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
tensor. If is of type `list[ImageInput]`, it's assumed that this is for a single prompt i.e. of batch size 1.
text (`Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]]`, *optional*):
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).
Wherever an image token, `` is encountered it is expanded to
`` + `` * `image_seq_len` * `.
return_tensors (`Union[str, TensorType]`, *optional*):
If set, will return tensors of a particular framework. See [`PreTrainedTokenizerFast.__call__`] for more
information.
Nz+You must provide either `text` or `images`.tokenizer_init_kwargsr2 r% r6 return_tensorsr zAInvalid input text. Please provide a string, or a list of strings image)
modalitieszThe total number of zP tokens in the prompts should be the same as the number of images passed. Found z tokens and z images. zdInvalid input images. Please provide a single image or a list of images or a list of list of images.z!The number of images in the text z
and images z should be the same.) tensor_type)rE
_merge_kwargsr0 r< init_kwargspopr% r r listrH rA r; do_image_splittingrU countreplace_check_special_mm_tokensupdater! tuplesumlenr ranger r )rO rY r\ audiovideosrP
output_kwargsr% r_ n_images_in_textinputsrH rA image_strprompt_stringssample text_inputscumsum_images_in_textin_images_in_imagesim image_inputss r __call__Idefics2Processor.__call__ s l N(N 78H7I WiVjj~
GMMfF7"z"~7fFM //Y-:XYL MM, 'F??7 "D 8Ms$ 2M:.M?$N ,N N N c : U R R " U0 UD6$ )z
This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
refer to the docstring of this method for more information.
)r< batch_decoderO argsrP s r r Idefics2Processor.batch_decode s
~~**D;F;;r c : U R R " U0 UD6$ )z
This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
the docstring of this method for more information.
)r< decoder s r r Idefics2Processor.decode s
~~$$d5f55r c U R R nU R R n[ [ R X- 5 5 $ r )r< model_input_namesr; ri dictfromkeys)rO tokenizer_input_namesimage_processor_input_namess r r #Idefics2Processor.model_input_names s< $ @ @ &*&:&:&L&L#DMM"7"UVWWr )rL rH r% rA rJ )N@ N)NNNN)r' r( r) r* __doc__
attributesimage_processor_classtokenizer_classr+ r r rN rZ r r ri r r r0 r r r r propertyr r-
__classcell__)rR s @r r9 r9 B s & $[1J 4%O hlR>ARW_`cWdR R2
OSbf H@j$z"2Dj9I4JJKH@ I 2DOTJ]E^^_H@ 01
H@
H@T<6 X Xr r9 N)"r itertoolsr typingr r r feature_extraction_utilsr image_utilsr r
r processing_utilsr r
r r tokenization_utils_baser r utilsr r
get_loggerr' loggerboolr r! r# r0 r9 __all__r& r r r s ! 1 1 4 A A = <
H %;4 ;0!