Add Idefics 3!

docs/source/en/_toctree.yml

@@ -820,6 +820,8 @@
         title: IDEFICS
       - local: model_doc/idefics2
         title: Idefics2
+      - local: model_doc/idefics3
+        title: Idefics3
       - local: model_doc/instructblip
         title: InstructBLIP
       - local: model_doc/instructblipvideo

docs/source/en/index.md

@@ -168,6 +168,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                        [I-BERT](model_doc/ibert)                         |       ✅        |         ❌         |      ❌      |
 |                       [IDEFICS](model_doc/idefics)                       |       ✅        |         ✅         |      ❌      |
 |                      [Idefics2](model_doc/idefics2)                      |       ✅        |         ❌         |      ❌      |
+|                      [Idefics3](model_doc/idefics3)                      |       ✅        |         ❌         |      ❌      |
 |                      [ImageGPT](model_doc/imagegpt)                      |       ✅        |         ❌         |      ❌      |
 |                      [Informer](model_doc/informer)                      |       ✅        |         ❌         |      ❌      |
 |                  [InstructBLIP](model_doc/instructblip)                  |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/idefics3.md

@@ -0,0 +1,73 @@
+<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+
+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.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Idefics3
+
+## Overview
+
+The Idefics3 model was proposed in [Building and better understanding vision-language models: insights and future directions](https://huggingface.co/papers/2408.12637) by Hugo Laurençon, Andrés Marafioti, Victor Sanh, and Léo Tronchon.
+
+Idefics3 is an adaptation of the Idefics2 model with three main differences:
+
+- It uses Llama3 for the text model.
+- It uses an updated processing logic for the images.
+- It removes the perceiver.
+
+The abstract from the paper is the following:
+
+*The field of vision-language models (VLMs), which take images and texts as inputs and output texts, is rapidly evolving and has yet to reach consensus on several key aspects of the development pipeline, including data, architecture, and training methods. This paper can be seen as a tutorial for building a VLM. We begin by providing a comprehensive overview of the current state-of-the-art approaches, highlighting the strengths and weaknesses of each, addressing the major challenges in the field, and suggesting promising research directions for underexplored areas. We then walk through the practical steps to build Idefics3-8B, a powerful VLM that significantly outperforms its predecessor Idefics2-8B, while being trained efficiently, exclusively on open datasets, and using a straightforward pipeline. These steps include the creation of Docmatix, a dataset for improving document understanding capabilities, which is 240 times larger than previously available datasets. We release the model along with the datasets created for its training.*
+
+## Usage tips
+
+Input images are processed either by upsampling (if resizing is enabled) or at their original resolution. The resizing behavior depends on two parameters: do_resize and size.
+
+If `do_resize` is set to `True`, the model resizes images so that the longest edge is 4*364 pixels by default.
+The default resizing behavior can be customized by passing a dictionary to the `size` parameter. For example, `{"longest_edge": 4 * 364}` is the default, but you can change it to a different value if needed.
+
+Here’s how to control resizing and set a custom size:
+```python
+image_processor = Idefics3ImageProcessor(do_resize=True, size={"longest_edge": 2 * 364}, max_image_size=364)
+```
+
+Additionally, the `max_image_size` parameter, which controls the size of each square patch the image is decomposed into, is set to 364 by default but can be adjusted as needed. After resizing (if applicable), the image processor decomposes the images into square patches based on the `max_image_size` parameter.
+
+This model was contributed by [amyeroberts](https://huggingface.co/amyeroberts) and [andimarafioti](https://huggingface.co/andito).
+
+
+## Idefics3Config
+
+[[autodoc]] Idefics3Config
+
+
+## Idefics3Model
+
+[[autodoc]] Idefics3Model
+    - forward
+
+## Idefics3ForConditionalGeneration
+
+[[autodoc]] Idefics3ForConditionalGeneration
+    - forward
+
+
+## Idefics3ImageProcessor
+[[autodoc]] Idefics3ImageProcessor
+    - preprocess
+
+
+## Idefics3Processor
+[[autodoc]] Idefics3Processor
+    - __call__

docs/source/en/perf_infer_gpu_one.md

@@ -53,6 +53,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj#transformers.GPTJModel)
 * [Granite](https://huggingface.co/docs/transformers/model_doc/granite#transformers.GraniteModel)
 * [Idefics2](https://huggingface.co/docs/transformers/model_doc/idefics2#transformers.Idefics2Model)
+* [Idefics3](https://huggingface.co/docs/transformers/model_doc/idefics3#transformers.Idefics3Model)
 * [Falcon](https://huggingface.co/docs/transformers/model_doc/falcon#transformers.FalconModel)
 * [JetMoe](https://huggingface.co/docs/transformers/model_doc/jetmoe#transformers.JetMoeModel)
 * [Jamba](https://huggingface.co/docs/transformers/model_doc/jamba#transformers.JambaModel)

src/transformers/__init__.py

@@ -480,6 +480,7 @@
     "models.ibert": ["IBertConfig"],
     "models.idefics": ["IdeficsConfig"],
     "models.idefics2": ["Idefics2Config"],
+    "models.idefics3": ["Idefics3Config"],
     "models.imagegpt": ["ImageGPTConfig"],
     "models.informer": ["InformerConfig"],
     "models.instructblip": [
@@ -1180,6 +1181,7 @@
     _import_structure["models.grounding_dino"].extend(["GroundingDinoImageProcessor"])
     _import_structure["models.idefics"].extend(["IdeficsImageProcessor"])
     _import_structure["models.idefics2"].extend(["Idefics2ImageProcessor"])
+    _import_structure["models.idefics3"].extend(["Idefics3ImageProcessor"])
     _import_structure["models.imagegpt"].extend(["ImageGPTFeatureExtractor", "ImageGPTImageProcessor"])
     _import_structure["models.instructblipvideo"].extend(["InstructBlipVideoImageProcessor"])
     _import_structure["models.layoutlmv2"].extend(["LayoutLMv2FeatureExtractor", "LayoutLMv2ImageProcessor"])
@@ -2401,6 +2403,14 @@
             "Idefics2Processor",
         ]
     )
+    _import_structure["models.idefics3"].extend(
+        [
+            "Idefics3ForConditionalGeneration",
+            "Idefics3Model",
+            "Idefics3PreTrainedModel",
+            "Idefics3Processor",
+        ]
+    )
     _import_structure["models.imagegpt"].extend(
         [
             "ImageGPTForCausalImageModeling",
@@ -5247,6 +5257,7 @@
         IdeficsConfig,
     )
     from .models.idefics2 import Idefics2Config
+    from .models.idefics3 import Idefics3Config
     from .models.imagegpt import ImageGPTConfig
     from .models.informer import InformerConfig
     from .models.instructblip import (
@@ -5983,6 +5994,7 @@
         from .models.grounding_dino import GroundingDinoImageProcessor
         from .models.idefics import IdeficsImageProcessor
         from .models.idefics2 import Idefics2ImageProcessor
+        from .models.idefics3 import Idefics3ImageProcessor
         from .models.imagegpt import ImageGPTFeatureExtractor, ImageGPTImageProcessor
         from .models.instructblipvideo import InstructBlipVideoImageProcessor
         from .models.layoutlmv2 import (
@@ -7011,6 +7023,12 @@
             Idefics2PreTrainedModel,
             Idefics2Processor,
         )
+        from .models.idefics3 import (
+            Idefics3ForConditionalGeneration,
+            Idefics3Model,
+            Idefics3PreTrainedModel,
+            Idefics3Processor,
+        )
         from .models.imagegpt import (
             ImageGPTForCausalImageModeling,
             ImageGPTForImageClassification,

src/transformers/image_transforms.py

@@ -162,6 +162,7 @@ def _rescale_for_pil_conversion(image):
 def to_pil_image(
     image: Union[np.ndarray, "PIL.Image.Image", "torch.Tensor", "tf.Tensor", "jnp.ndarray"],
     do_rescale: Optional[bool] = None,
+    image_mode: Optional[str] = None,
     input_data_format: Optional[Union[str, ChannelDimension]] = None,
 ) -> "PIL.Image.Image":
     """
@@ -175,6 +176,8 @@ def to_pil_image(
             Whether or not to apply the scaling factor (to make pixel values integers between 0 and 255). Will default
             to `True` if the image type is a floating type and casting to `int` would result in a loss of precision,
             and `False` otherwise.
+        image_mode (`str`, *optional*):
+            The mode to use for the PIL image. If unset, will use the default mode for the input image type.
         input_data_format (`ChannelDimension`, *optional*):
             The channel dimension format of the input image. If unset, will use the inferred format from the input.
 
@@ -207,7 +210,7 @@ def to_pil_image(
         image = rescale(image, 255)
 
     image = image.astype(np.uint8)
-    return PIL.Image.fromarray(image)
+    return PIL.Image.fromarray(image, mode=image_mode)
 
 
 # Logic adapted from torchvision resizing logic: https://github.com/pytorch/vision/blob/511924c1ced4ce0461197e5caa64ce5b9e558aab/torchvision/transforms/functional.py#L366

src/transformers/models/__init__.py

@@ -114,6 +114,7 @@
     ibert,
     idefics,
     idefics2,
+    idefics3,
     imagegpt,
     informer,
     instructblip,

src/transformers/models/auto/configuration_auto.py

@@ -131,6 +131,7 @@
         ("ibert", "IBertConfig"),
         ("idefics", "IdeficsConfig"),
         ("idefics2", "Idefics2Config"),
+        ("idefics3", "Idefics3Config"),
         ("imagegpt", "ImageGPTConfig"),
         ("informer", "InformerConfig"),
         ("instructblip", "InstructBlipConfig"),
@@ -425,6 +426,7 @@
         ("ibert", "I-BERT"),
         ("idefics", "IDEFICS"),
         ("idefics2", "Idefics2"),
+        ("idefics3", "Idefics3"),
         ("imagegpt", "ImageGPT"),
         ("informer", "Informer"),
         ("instructblip", "InstructBLIP"),

src/transformers/models/auto/image_processing_auto.py

@@ -89,6 +89,7 @@
             ("hiera", ("BitImageProcessor",)),
             ("idefics", ("IdeficsImageProcessor",)),
             ("idefics2", ("Idefics2ImageProcessor",)),
+            ("idefics3", ("Idefics3ImageProcessor",)),
             ("imagegpt", ("ImageGPTImageProcessor",)),
             ("instructblip", ("BlipImageProcessor",)),
             ("instructblipvideo", ("InstructBlipVideoImageProcessor",)),

src/transformers/models/auto/modeling_auto.py

@@ -128,6 +128,7 @@
         ("ibert", "IBertModel"),
         ("idefics", "IdeficsModel"),
         ("idefics2", "Idefics2Model"),
+        ("idefics3", "Idefics3Model"),
         ("imagegpt", "ImageGPTModel"),
         ("informer", "InformerModel"),
         ("jamba", "JambaModel"),
@@ -311,6 +312,7 @@
         ("ibert", "IBertForMaskedLM"),
         ("idefics", "IdeficsForVisionText2Text"),
         ("idefics2", "Idefics2ForConditionalGeneration"),
+        ("idefics3", "Idefics3ForConditionalGeneration"),
         ("layoutlm", "LayoutLMForMaskedLM"),
         ("llava", "LlavaForConditionalGeneration"),
         ("llava_next", "LlavaNextForConditionalGeneration"),
@@ -725,6 +727,7 @@
         ("chameleon", "ChameleonForConditionalGeneration"),
         ("git", "GitForCausalLM"),
         ("idefics2", "Idefics2ForConditionalGeneration"),
+        ("idefics3", "Idefics3ForConditionalGeneration"),
         ("instructblip", "InstructBlipForConditionalGeneration"),
         ("instructblipvideo", "InstructBlipVideoForConditionalGeneration"),
         ("kosmos-2", "Kosmos2ForConditionalGeneration"),

src/transformers/models/auto/processing_auto.py

@@ -65,6 +65,7 @@
         ("hubert", "Wav2Vec2Processor"),
         ("idefics", "IdeficsProcessor"),
         ("idefics2", "Idefics2Processor"),
+        ("idefics3", "Idefics3Processor"),
         ("instructblip", "InstructBlipProcessor"),
         ("instructblipvideo", "InstructBlipVideoProcessor"),
         ("kosmos-2", "Kosmos2Processor"),

src/transformers/models/auto/tokenization_auto.py

@@ -219,6 +219,7 @@
             ("ibert", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)),
             ("idefics", (None, "LlamaTokenizerFast" if is_tokenizers_available() else None)),
             ("idefics2", ("LlamaTokenizer", "LlamaTokenizerFast" if is_tokenizers_available() else None)),
+            ("idefics3", ("LlamaTokenizer", "LlamaTokenizerFast" if is_tokenizers_available() else None)),
             ("instructblip", ("GPT2Tokenizer", "GPT2TokenizerFast" if is_tokenizers_available() else None)),
             ("instructblipvideo", ("GPT2Tokenizer", "GPT2TokenizerFast" if is_tokenizers_available() else None)),
             (

src/transformers/models/idefics2/modeling_idefics2.py

@@ -1097,7 +1097,7 @@ class Idefics2PreTrainedModel(PreTrainedModel):
 
     def _init_weights(self, module):
         std = (
-            self.config.text_config.initializer_range
+            self.config.initializer_range
             if hasattr(self.config, "initializer_range")
             else self.config.text_config.initializer_range
         )

src/transformers/models/idefics3/__init__.py

@@ -0,0 +1,72 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# 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.
+from typing import TYPE_CHECKING
+
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available, is_vision_available
+
+
+_import_structure = {"configuration_idefics3": ["Idefics3Config"]}
+
+
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["image_processing_idefics3"] = ["Idefics3ImageProcessor"]
+
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_idefics3"] = [
+        "Idefics3ForConditionalGeneration",
+        "Idefics3PreTrainedModel",
+        "Idefics3Model",
+    ]
+    _import_structure["processing_idefics3"] = ["Idefics3Processor"]
+
+if TYPE_CHECKING:
+    from .configuration_idefics3 import Idefics3Config
+
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .image_processing_idefics3 import Idefics3ImageProcessor
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_idefics3 import (
+            Idefics3ForConditionalGeneration,
+            Idefics3Model,
+            Idefics3PreTrainedModel,
+        )
+        from .processing_idefics3 import Idefics3Processor
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure)