Hugging Face's logo

Transformers documentation

IDEFICS

You are viewing v4.40.2 version. A newer version v4.46.3 is available.
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

IDEFICS

Overview

The IDEFICS model was proposed in OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents by Hugo Laurençon, Lucile Saulnier, Léo Tronchon, Stas Bekman, Amanpreet Singh, Anton Lozhkov, Thomas Wang, Siddharth Karamcheti, Alexander M. Rush, Douwe Kiela, Matthieu Cord, Victor Sanh

The abstract from the paper is the following:

Large multimodal models trained on natural documents, which interleave images and text, outperform models trained on image-text pairs on various multimodal benchmarks that require reasoning over one or multiple images to generate a text. However, the datasets used to train these models have not been released, and the collection process has not been fully specified. We introduce the OBELICS dataset, an open web-scale filtered dataset of interleaved image-text documents comprising 141 million web pages extracted from Common Crawl, 353 million associated images, and 115 billion text tokens. We describe the dataset creation process, present comprehensive filtering rules, and provide an analysis of the dataset’s content. To show the viability of OBELISC, we train an 80 billion parameters vision and language model on the dataset and obtain competitive performance on various multimodal benchmarks. We release the code to reproduce the dataset along with the dataset itself.

This model was contributed by HuggingFaceM4. The original code can be found here. (TODO: don’t have a public link yet).

IDEFICS modeling code in Transformers is for finetuning and inferencing the pre-trained IDEFICS models.

To train a new IDEFICS model from scratch use the m4 codebase (a link will be provided once it’s made public)

IdeficsConfig

class transformers.IdeficsConfig

< >

( vocab_size = 32000 additional_vocab_size = 0 hidden_size = 4096 intermediate_size = 11008 num_hidden_layers = 32 num_attention_heads = 32 dropout = 0.0 hidden_act = 'silu' initializer_range = 0.02 alpha_initializer = 'zeros' alphas_initializer_range = 0.0 alpha_type = 'float' rms_norm_eps = 1e-06 use_cache = True pad_token_id = 0 bos_token_id = 1 eos_token_id = 2 tie_word_embeddings = False cross_layer_interval = 1 qk_layer_norms = False freeze_text_layers = True freeze_text_module_exceptions = [] freeze_lm_head = False freeze_vision_layers = True freeze_vision_module_exceptions = [] use_resampler = False vision_config = None perceiver_config = None **kwargs )

Parameters

  • additional_vocab_size (int, *optional`, defaults to 0) — Additional vocabulary size of the model, typically for the special ”” token. Additional vocab tokens are always trainable whereas regular vocab tokens can be frozen or not.
  • vocab_size (int, optional, defaults to 32000) — Vocabulary size of the Idefics model. Defines the number of different tokens that can be represented by the inputs_ids passed when calling ~IdeficsModel
  • hidden_size (int, optional, defaults to 4096) — Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 11008) — Dimension of the MLP representations.
  • num_hidden_layers (int, optional, defaults to 32) — Number of hidden layers in the Transformer encoder.
  • num_attention_heads (int, optional, defaults to 32) — Number of attention heads for each attention layer in the Transformer encoder.
  • dropout (float, optional, defaults to 0.0) — The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
  • hidden_act (str or function, optional, defaults to "silu") — The non-linear activation function (function or string) in the decoder.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • alpha_initializer (str, optional, defaults to "zeros") — Initialization type for the alphas.
  • alphas_initializer_range (float, optional, defaults to 0.0) — The standard deviation of the truncated_normal_initializer for initializing the alphas in the Gated Cross Attention.
  • alpha_type (str, optional, defaults to "float") — Whether the gating alphas should be vectors or single floats.
  • rms_norm_eps (float, optional, defaults to 1e-6) — The epsilon used by the rms normalization layers.
  • use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True.
  • pad_token_id (int, optional, defaults to 0) — Padding token id.
  • bos_token_id (int, optional, defaults to 1) — Beginning of stream token id.
  • eos_token_id (int, optional, defaults to 2) — End of stream token id.
  • tie_word_embeddings(bool, optional, defaults to False) — Whether to tie weight embeddings
  • cross_layer_interval (int, optional, default to 1) — Interval for cross attention (from text to image) layers.
  • qk_layer_norms (bool, optional, defaults to False) — Whether to add layer norm after q and k
  • freeze_text_layers (bool, optional, defaults to True) — Whether to freeze text layers
  • freeze_text_module_exceptions (bool, optional, defaults to []) — Exceptions to freezing text layers when freeze_text_layers is True
  • freeze_lm_head (bool, optional, defaults to False) — Whether to freeze lm head
  • freeze_vision_layers (bool, optional, defaults to True) — Whether to freeze vision layers
  • freeze_vision_module_exceptions (bool, optional, defaults to []) — Exceptions to freezing vision layers when freeze_vision_layers is True
  • use_resampler (bool, optional, defaults to False) — Whether to use the Resampler
  • vision_config (IdeficsVisionConfig, optional) — Custom vision config or dict
  • perceiver_config (IdeficsPerceiverConfig, optional) — Custom perceiver config or dict

This is the configuration class to store the configuration of a IdeficsModel. It is used to instantiate an Idefics model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the Idefics-9B.

e.g. HuggingFaceM4/idefics-9b

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

Example:

>>> from transformers import IdeficsModel, IdeficsConfig

>>> # Initializing a Idefics idefics-9b style configuration
>>> configuration = IdeficsConfig()

>>> # Initializing a model from the idefics-9b style configuration
>>> model = IdeficsModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

IdeficsModel

class transformers.IdeficsModel

< >

( config: IdeficsConfig )

Parameters

  • config (IdeficsConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights. config — IdeficsConfig

The bare LLaMA Model outputting raw hidden-states without any specific head on top. This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)

This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

Transformer decoder consisting of config.num_hidden_layers layers. Each layer is a IdeficsDecoderLayer

forward

< >

( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None pixel_values: Optional = None image_encoder_embeddings: Optional = None perceiver_embeddings: Optional = None image_attention_mask: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None interpolate_pos_encoding: Optional = False return_dict: Optional = None )

Parameters

  • input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide it.

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    What are input IDs?

  • attention_mask (torch.Tensor of shape (batch_size, sequence_length), optional) — Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:

    • 1 for tokens that are not masked,
    • 0 for tokens that are masked.

    What are attention masks?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last decoder_input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1]. What are position IDs?
  • past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape (batch_size, num_heads, encoder_sequence_length, embed_size_per_head).

    Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

    If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all decoder_input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
  • return_dict (bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.

The IdeficsModel forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

IdeficsForVisionText2Text

class transformers.IdeficsForVisionText2Text

< >

( config vision_model = None )

forward

< >

( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None pixel_values: Optional = None image_encoder_embeddings: Optional = None perceiver_embeddings: Optional = None image_attention_mask: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None interpolate_pos_encoding: Optional = False return_dict: Optional = None ) → transformers.models.idefics.modeling_idefics.IdeficsCausalLMOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • input_ids (torch.LongTensor of shape (batch_size, sequence_length)) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide it.

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    What are input IDs?

  • attention_mask (torch.Tensor of shape (batch_size, sequence_length), optional) — Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:

    • 1 for tokens that are not masked,
    • 0 for tokens that are masked.

    What are attention masks?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last decoder_input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1]. What are position IDs?
  • past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)) and 2 additional tensors of shape (batch_size, num_heads, encoder_sequence_length, embed_size_per_head).

    Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

    If past_key_values are used, the user can optionally input only the last decoder_input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all decoder_input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for more detail.
  • return_dict (bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.

    Args — 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 -100 (see input_ids docstring). Tokens with indices set to -100 are ignored (masked), the loss is only computed for the tokens with labels in [0, ..., config.vocab_size].

Returns

transformers.models.idefics.modeling_idefics.IdeficsCausalLMOutputWithPast or tuple(torch.FloatTensor)

A transformers.models.idefics.modeling_idefics.IdeficsCausalLMOutputWithPast or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (IdeficsConfig) and inputs.

  • loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction).

  • logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

  • past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head))

    Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

  • hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

    Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.

  • attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

    Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

  • image_hidden_states (tuple(torch.FloatTensor), optional) — Tuple of torch.FloatTensor (one for the output of the image embeddings, (batch_size, num_images, sequence_length, hidden_size).

    image_hidden_states of the model produced by the vision encoder, and optionally by the perceiver

The IdeficsForVisionText2Text forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

Example:

>>> from transformers import AutoProcessor, IdeficsForVisionText2Text

>>> model = IdeficsForVisionText2Text.from_pretrained("HuggingFaceM4/idefics-9b")
>>> processor = AutoProcessor.from_pretrained("HuggingFaceM4/idefics-9b")

>>> dogs_image_url_1 = "https://huggingface.co/datasets/hf-internal-testing/fixtures_nlvr2/raw/main/image1.jpeg"
>>> dogs_image_url_2 = "https://huggingface.co/datasets/hf-internal-testing/fixtures_nlvr2/raw/main/image2.jpeg"

>>> prompts = [
...     [
...         "User:",
...         dogs_image_url_1,
...         "Describe this image.\nAssistant: An image of two dogs.\n",
...         "User:",
...         dogs_image_url_2,
...         "Describe this image.\nAssistant:",
...     ]
... ]
>>> inputs = processor(prompts, return_tensors="pt")
>>> generate_ids = model.generate(**inputs, max_new_tokens=6)
>>> processor.batch_decode(generate_ids, skip_special_tokens=True)

IdeficsImageProcessor

class transformers.IdeficsImageProcessor

< >

( image_size: int = 224 image_mean: Union = None image_std: Union = None image_num_channels: Optional = 3 **kwargs )

Parameters

  • image_size (int, optional, defaults to 224) — Resize to image size
  • image_mean (float or List[float], optional, defaults to IDEFICS_STANDARD_MEAN) — Mean to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the image_mean parameter in the preprocess method. Can be overridden by the image_mean parameter in the preprocess method.
  • image_std (float or List[float], optional, defaults to IDEFICS_STANDARD_STD) — Standard deviation to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the image_std parameter in the preprocess method. Can be overridden by the image_std parameter in the preprocess method.
  • image_num_channels (int, optional, defaults to 3) — Number of image channels.

Constructs a Idefics image processor.

preprocess

< >

( images: Union image_num_channels: Optional = 3 image_size: Optional = None image_mean: Union = None image_std: Union = None transform: Callable = None **kwargs )

Parameters

  • images (ImageInput) — A list of images to preprocess.
  • image_size (int, optional, defaults to self.image_size) — Resize to image size
  • image_num_channels (int, optional, defaults to self.image_num_channels) — Number of image channels.
  • image_mean (float or List[float], optional, defaults to IDEFICS_STANDARD_MEAN) — Mean to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the image_mean parameter in the preprocess method. Can be overridden by the image_mean parameter in the preprocess method.
  • image_std (float or List[float], optional, defaults to IDEFICS_STANDARD_STD) — Standard deviation to use if normalizing the image. This is a float or list of floats the length of the number of channels in the image. Can be overridden by the image_std parameter in the preprocess method. Can be overridden by the image_std parameter in the preprocess method.
  • transform (Callable, optional, defaults to None) — A custom transform function that accepts a single image can be passed for training. For example, torchvision.Compose can be used to compose multiple transforms. If None - an inference mode is assumed - and then a preset of inference-specific transforms will be applied to the images

Preprocess a batch of images.

IdeficsProcessor

class transformers.IdeficsProcessor

< >

( image_processor tokenizer = None image_size = 224 add_end_of_utterance_token = None **kwargs )

Parameters

  • image_processor (IdeficsImageProcessor) — An instance of IdeficsImageProcessor. The image processor is a required input.
  • tokenizer (LlamaTokenizerFast) — An instance of LlamaTokenizerFast. The tokenizer is a required input.
  • image_size (int, optional, defaults to 224) — Image size (assuming a square image)

Constructs a IDEFICS processor which wraps a LLama tokenizer and IDEFICS image processor into a single processor.

IdeficsProcessor offers all the functionalities of IdeficsImageProcessor and LlamaTokenizerFast. See the docstring of call() and decode() for more information.

__call__

< >

( prompts: Union padding: Union = 'longest' truncation: Union = None max_length: Optional = None transform: Callable = None add_eos_token = False add_end_of_utterance_token = None debug = False return_tensors: Union = <TensorType.PYTORCH: 'pt'> ) → a dict with entries

Parameters

  • prompts (Union[List[TextInput], [List[List[TextInput]]]]) — either a single prompt or a batched list of prompts - see the detailed description immediately after the end of the arguments doc section.
  • padding (bool, str or PaddingStrategy, optional, defaults to "longest") — Select a strategy to pad the returned sequences (according to the model’s padding side and padding index) among:
    • True or 'longest' (default): Pad to the longest sequence in the batch (or no padding if only a single sequence if provided).
    • 'max_length': Pad to a maximum length specified with the argument max_length or to the maximum acceptable input length for the model if that argument is not provided.
    • False or 'do_not_pad': No padding. This will raise an error if the input sequences are of different lengths. Note: Unlike most processors, which set padding=False by default, IdeficsProcessor sets padding="longest" by default. See https://github.com/huggingface/transformers/pull/29449#pullrequestreview-1925576061 for why.
  • max_length (int, optional) — Maximum length of the returned list and optionally padding length (see above).
  • truncation (bool, optional) — Activates truncation to cut input sequences longer than max_length to max_length.
  • transform (Callable, optional) — A custom transform function that accepts a single image can be passed for training. For example, torchvision.Compose can be used to compose multiple functions. If None a preset inference-specific set of transforms will be applied to the images
  • add_eos_token (bool, optional, defaults to False) — Adds eos_token at the end of the final prompt if True`
  • add_end_of_utterance_token (bool, optional) — Whether to automatically add <end_of_utterance> after each prompt’s text input (unless followed by an image). If None the tokenizer will be checked instead and if this token is found in additional_special_tokens then the value will be True.
  • debug (bool, optional, defaults to False) — True value will help debug prompt generation by dumping useful information
  • return_tensors (str or TensorType, optional, defaults to TensorType.PYTORCH) — The type of tensors to return. Can be one of:
    • TensorType.PYTORCH or 'pt': Return a batch of type torch.Tensor.

Returns

a dict with entries

input_ids, attention_mask, pixel_values, image_attention_mask which can be directly passed to model.generate

This method takes batched or non-batched prompts made of text and images and converts them into prompts that the model was trained on and prepares the image pixel values for the model to process.

Detailed explanation:

Each entry in prompts is either a text to be passed as is or an image that will be processed.

An image can be either an image object (PIL.Image) or a url from which the image can be retrieved.

When the processor encounters an image it’ll inject <fake_token_around_image><image><fake_token_around_image> entry into the prompt.

Example:

checkpoint = "HuggingFaceM4/idefics-9b"
processor = AutoProcessor.from_pretrained(checkpoint)
url = "https://hips.hearstapps.com/hmg-prod/images/cute-photos-of-cats-in-grass-1593184777.jpg"
img = processor.image_processor.fetch_images([url])[0]

prompts = [
    "User:",
    img,
    "Describe this image.
t: An image of two kittens in grass.

    "User:",
    "https://hips.hearstapps.com/hmg-prod/images/dog-puns-1581708208.jpg",
    "Describe this image.
t:",
]

inputs = processor(prompts, return_tensors="pt")
generated_ids = model.generate(**inputs, max_length=100)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

In this example the prompts will be converted into:

<s>User:<fake_token_around_image><image><fake_token_around_image>Describe this image.
Assistant: An image of two kittens in grass.
User:<fake_token_around_image><image><fake_token_around_image>Describe this image.
Assistant:'

and the two images will be massaged using IdeficsImageProcessor.call() method and placed inside the pixel_values dict entry of the return value.

This example also examplifies that images can be passed as objects or as text urls. It can be seen that the first image is passed as object and the second one as a url.

To do training do:

image_transform = transforms.Compose(
    [
        transforms.RandomResizedCrop(
            (w, h), scale=(0.9, 1.0), interpolation=transforms.InterpolationMode.BICUBIC
        ),
        transforms.ToTensor(),
        transforms.Normalize(mean=self.image_mean, std=self.image_std),
    ]
)
inputs = processor(prompts, transform=image_transform, return_tensors="pt")

In order to help debug prompt generation enable debug=True which will show you what’s happening.

< > Update on GitHub

←GroupViT Idefics2→