# coding=utf-8
# Copyright 2024 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Processor class for IDEFICS2.
"""
from itertools import accumulate
from typing import TYPE_CHECKING, List, Optional, Union
from ...feature_extraction_utils import BatchFeature
from ...image_utils import ImageInput, is_valid_image, load_image
from ...processing_utils import (
ImagesKwargs,
ProcessingKwargs,
ProcessorMixin,
Unpack,
_validate_images_text_input_order,
)
from ...tokenization_utils_base import AddedToken, TextInput
from ...utils import logging
if TYPE_CHECKING:
from ...tokenization_utils_base import PreTokenizedInput
logger = logging.get_logger(__name__)
def is_url(val) -> bool:
return isinstance(val, str) and val.startswith("http")
def is_image_or_image_url(elem):
return is_url(elem) or is_valid_image(elem)
class Idefics2ImagesKwargs(ImagesKwargs, total=False):
image_seq_len: Optional[int]
class Idefics2ProcessorKwargs(ProcessingKwargs, total=False):
images_kwargs: Idefics2ImagesKwargs
_defaults = {
"text_kwargs": {
"add_special_tokens": True,
"padding": False,
"is_split_into_words": False,
},
"images_kwargs": {},
}
class Idefics2Processor(ProcessorMixin):
r"""
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 <image> 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.
"""
attributes = ["image_processor", "tokenizer"]
valid_kwargs = ["image_seq_len", "chat_template"]
image_processor_class = "Idefics2ImageProcessor"
tokenizer_class = "AutoTokenizer"
def __init__(
self, image_processor, tokenizer=None, image_seq_len: int = 64, chat_template: Optional[str] = None, **kwargs
):
if image_processor is None:
raise ValueError("You need to specify an `image_processor`.")
if tokenizer is None:
raise ValueError("You need to specify a `tokenizer`.")
if not hasattr(tokenizer, "image_token"):
self.fake_image_token = AddedToken("<fake_token_around_image>", normalized=False, special=True)
self.image_token = AddedToken("<image>", normalized=False, special=True)
tokens_to_add = {"additional_special_tokens": [self.fake_image_token, self.image_token]}
tokenizer.add_special_tokens(tokens_to_add)
else:
self.fake_image_token = tokenizer.image_boundary_token
self.image_token = tokenizer.image_token
self.end_of_utterance_token = AddedToken("<end_of_utterance>", normalized=False, special=True)
tokenizer.add_special_tokens({"additional_special_tokens": [self.end_of_utterance_token]})
self.image_seq_len = image_seq_len
super().__init__(image_processor, tokenizer, chat_template=chat_template)
def _extract_images_from_prompts(self, prompts):
prompt_images = []
for prompt in prompts:
images = []
for elem in prompt:
if is_valid_image(elem):
images.append(elem)
elif is_url(elem):
images.append(load_image(elem))
prompt_images.append(images)
return prompt_images
def __call__(
self,
images: Union[ImageInput, List[ImageInput], List[List[ImageInput]]] = None,
text: Union[TextInput, "PreTokenizedInput", List[TextInput], List["PreTokenizedInput"]] = None,
audio=None,
videos=None,
**kwargs: Unpack[Idefics2ProcessorKwargs],
) -> BatchFeature:
"""
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 = [
... "<image>In this image, we see",
... "bla bla bla<image>",
... ]
>>> 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)
['<s><fake_token_around_image><image><image><fake_token_around_image> In this image, we see', '<s> bla bla bla<fake_token_around_image><image><image><fake_token_around_image>']
```
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, `<image>` is encountered it is expanded to
`<fake_token_around_image>` + `<image>` * `image_seq_len` * <fake_token_around_image>`.
return_tensors (`Union[str, TensorType]`, *optional*):
If set, will return tensors of a particular framework. See [`PreTrainedTokenizerFast.__call__`] for more
information.
"""
if text is None and images is None:
raise ValueError("You must provide either `text` or `images`.")
# check if images and text inputs are reversed for BC
images, text = _validate_images_text_input_order(images, text)
output_kwargs = self._merge_kwargs(
Idefics2ProcessorKwargs,
tokenizer_init_kwargs=self.tokenizer.init_kwargs,
**kwargs,
)
image_seq_len = output_kwargs["images_kwargs"].pop("image_seq_len", None)
image_seq_len = image_seq_len if image_seq_len is not None else self.image_seq_len
n_images_in_text = []
inputs = BatchFeature()
if text is not None:
if isinstance(text, str):
text = [text]
elif not isinstance(text, list) and not isinstance(text[0], str):
raise ValueError("Invalid input text. Please provide a string, or a list of strings")
# Replace the image token with fake tokens around the expanded image token sequence of length `image_seq_len`
fake_image_token = self.fake_image_token.content
image_token = self.image_token.content
image_str = f"{fake_image_token}{image_token * image_seq_len}{fake_image_token}"
if self.image_processor.do_image_splitting:
# A single image token is split into 4 patches + 1 original image
image_str = image_str * 5
prompt_strings = []
for sample in text:
n_images_in_text.append(sample.count(image_token))
sample = sample.replace(image_token, image_str)
# Remove any double fake tokens if images are adjacent
sample = sample.replace(f"{fake_image_token}{fake_image_token}", f"{fake_image_token}")
prompt_strings.append(sample)
text_inputs = self.tokenizer(prompt_strings, **output_kwargs["text_kwargs"])
inputs.update(text_inputs)
if images is not None:
if is_image_or_image_url(images):
images = [[images]]
elif isinstance(images, (list, tuple)) and is_image_or_image_url(images[0]):
if text is not None:
if sum(n_images_in_text) != len(images):
raise ValueError(
f"The total number of {image_token} tokens in the prompts should be the same as the number of images passed."
f" Found {sum(n_images_in_text)} {image_token} tokens and {len(images)} images."
)
# Reorganize the images to match the prompts
cumsum_images_in_text = [0] + list(accumulate(n_images_in_text))
images = [
images[cumsum_images_in_text[i] : cumsum_images_in_text[i + 1]]
for i in range(len(n_images_in_text))
]
else:
images = [images]
elif (
not isinstance(images, (list, tuple))
and not isinstance(images[0], (list, tuple))
and not is_image_or_image_url(images[0][0])
):
raise ValueError(
"Invalid input images. Please provide a single image or a list of images or a list of list of images."
)
n_images_in_images = [len(sample) for sample in images]
if text is not None and not n_images_in_images == n_images_in_text:
raise ValueError(
f"The number of images in the text {n_images_in_text} and images {n_images_in_images} should be the same."
)
# Load images if they are URLs
images = [[load_image(im) for im in sample] for sample in images]
image_inputs = self.image_processor(images, **output_kwargs["images_kwargs"])
inputs.update(image_inputs)
return inputs
def batch_decode(self, *args, **kwargs):
"""
This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
refer to the docstring of this method for more information.
"""
return self.tokenizer.batch_decode(*args, **kwargs)
def decode(self, *args, **kwargs):
"""
This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
the docstring of this method for more information.
"""
return self.tokenizer.decode(*args, **kwargs)
@property
def model_input_names(self):
tokenizer_input_names = self.tokenizer.model_input_names
image_processor_input_names = self.image_processor.model_input_names
return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
__all__ = ["Idefics2Processor"]