Add Video Llava

README_fr.md

@@ -288,7 +288,6 @@ Suivez les pages d'installation de Flax, PyTorch ou TensorFlow pour voir comment
 
 Nombre actuel de points de contrôle : ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
-
 🤗 Transformers fournit actuellement les architectures suivantes: consultez [ici](https://huggingface.co/docs/transformers/model_summary) pour un résumé global de chacune d'entre elles.
 
 Pour vérifier si chaque modèle a une implémentation en Flax, PyTorch ou TensorFlow, ou s'il a un tokenizer associé pris en charge par la bibliothèque 🤗 Tokenizers, consultez [ce tableau](https://huggingface.co/docs/transformers/index#supported-frameworks).

README_te.md

@@ -293,7 +293,6 @@ Flax, PyTorch లేదా TensorFlow యొక్క ఇన్‌స్టా
 
 🤗 ట్రాన్స్‌ఫార్మర్లు ప్రస్తుతం కింది ఆర్కిటెక్చర్‌లను అందజేస్తున్నాయి: వాటిలో ప్రతి ఒక్కటి ఉన్నత స్థాయి సారాంశం కోసం [ఇక్కడ](https://huggingface.co/docs/transformers/model_summary) చూడండి.
 
-
 ఈ అమలులు అనేక డేటాసెట్‌లలో పరీక్షించబడ్డాయి (ఉదాహరణ స్క్రిప్ట్‌లను చూడండి) మరియు అసలైన అమలుల పనితీరుతో సరిపోలాలి. మీరు [డాక్యుమెంటేషన్](https://github.com/huggingface/transformers/tree/main/examples) యొక్క ఉదాహరణల విభాగంలో పనితీరుపై మరిన్ని వివరాలను కనుగొనవచ్చు.
 
 ## ఇంకా నేర్చుకో

docs/source/en/_toctree.yml

@@ -802,6 +802,8 @@
         title: TVP
       - local: model_doc/udop
         title: UDOP
+      - local: model_doc/video_llava
+        title: VideoLlava
       - local: model_doc/vilt
         title: ViLT
       - local: model_doc/vipllava

docs/source/en/index.md

@@ -302,6 +302,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                       [UnivNet](model_doc/univnet)                       |       ✅        |         ❌         |      ❌      |
 |                       [UPerNet](model_doc/upernet)                       |       ✅        |         ❌         |      ❌      |
 |                           [VAN](model_doc/van)                           |       ✅        |         ❌         |      ❌      |
+|                   [VideoLlava](model_doc/video_llava)                    |       ✅        |         ❌         |      ❌      |
 |                      [VideoMAE](model_doc/videomae)                      |       ✅        |         ❌         |      ❌      |
 |                          [ViLT](model_doc/vilt)                          |       ✅        |         ❌         |      ❌      |
 |                      [VipLlava](model_doc/vipllava)                      |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/video_llava.md

@@ -0,0 +1,129 @@
+<!--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.
+
+-->
+
+# Video-LLaVA
+
+## Overview
+
+Video-LLaVa is an open-source multimodal LLM trained by fine-tuning LlamA/Vicuna on multimodal instruction-following data generated by Llava1.5 and VideChat. It is an auto-regressive language model, based on the transformer architecture. Video-LLaVa unifies visual representations to the language feature space, and enables an LLM to perform visual reasoning capabilities on both images and videos simultaneously.
+
+
+The Video-LLaVA model was proposed in [Video-LLaVA: Learning United Visual Representation by Alignment Before Projection](https://arxiv.org/abs/2311.10122) by Bin Lin, Yang Ye, Bin Zhu, Jiaxi Cui, Munang Ning, Peng Jin, Li Yuan.
+
+The abstract from the paper is the following:
+
+*The Large Vision-Language Model (LVLM) has enhanced the performance of various downstream tasks in
+visual-language understanding. Most existing approaches
+encode images and videos into separate feature spaces,
+which are then fed as inputs to large language models.
+However, due to the lack of unified tokenization for images and videos, namely misalignment before projection, it
+becomes challenging for a Large Language Model (LLM)
+to learn multi-modal interactions from several poor projection layers. In this work, we unify visual representation into the language feature space to advance the foundational LLM towards a unified LVLM. As a result, we establish a simple but robust LVLM baseline, Video-LLaVA,
+which learns from a mixed dataset of images and videos,
+mutually enhancing each other. Video-LLaVA achieves superior performances on a broad range of 9 image benchmarks across 5 image question-answering datasets and 4
+image benchmark toolkits. Additionally, our Video-LLaVA
+also outperforms Video-ChatGPT by 5.8%, 9.9%, 18.6%,
+and 10.1% on MSRVTT, MSVD, TGIF, and ActivityNet, respectively. Notably, extensive experiments demonstrate that
+Video-LLaVA mutually benefits images and videos within
+a unified visual representation, outperforming models designed specifically for images or videos. We aim for this
+work to provide modest insights into the multi-modal inputs
+for the LLM*
+
+Tips:
+
+- We advise users to use padding_side="left" when computing batched generation as it leads to more accurate results. Simply make sure to call processor.tokenizer.padding_side = "left" before generating.
+
+- Note the model has not been explicitly trained to process multiple images/videos in the same prompt, although this is technically possible, you may experience inaccurate results.
+
+- For better results, we recommend users prompt the model with the correct prompt format:
+
+
+```python
+import av
+import torch
+import numpy as np
+import requests
+from PIL import Image
+from transformers import VideoLlavaForConditionalGeneration, VideoLlavaProcessor
+
+def read_video_pyav(container, indices):
+    '''
+    Decode the video with PyAV decoder.
+    Args:
+        container (`av.container.input.InputContainer`): PyAV container.
+        indices (`List[int]`): List of frame indices to decode.
+    Returns:
+        result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
+    '''
+    frames = []
+    container.seek(0)
+    start_index = indices[0]
+    end_index = indices[-1]
+    for i, frame in enumerate(container.decode(video=0)):
+        if i > end_index:
+            break
+        if i >= start_index and i in indices:
+            frames.append(frame)
+    return np.stack([x.to_ndarray(format="rgb24") for x in frames])
+
+
+model = VideoLlavaForConditionalGeneration.from_pretrained("RaushanTurganbay/video-llava-7b-hf", device_map="auto")
+processor = VideoLlavaProcessor.from_pretrained("RaushanTurganbay/video-llava-7b-hf")
+
+video_path = hf_hub_download(repo_id="raushan-testing-hf/videos-test", filename="sample_demo_1.mp4", repo_type="dataset")
+
+container = av.open(video_path)
+total_frames = container.streams.video[0].frames
+indices = np.arange(0, total_frames, total_frames / 8).astype(int)
+video = read_video_pyav(container, indices)
+
+prompt = "USER: <video>Why is this funny? ASSISTANT:"
+inputs = processor(text=prompt, videos=video, return_tensors="pt")
+
+out = model.generate(**inputs, max_new_tokens=40)
+print(processor.batch_decode(out, skip_special_tokens=True, clean_up_tokenization_spaces=True))
+```
+
+For multiple turns conversation change the prompt to:
+
+```bash
+"USER: <video>What do you see in this video? ASSISTANT: A baby reading a book. USER: Why is the it funny? ASSISTANT:"
+```
+
+- Note that the video inputs should have exactly 8 frames at the input, since the models were trained in that setting.
+
+
+
+This model was contributed by [RaushanTurganbay](https://huggingface.co/RaushanTurganbay).
+The original code can be found [here](https://github.com/PKU-YuanGroup/Video-LLaVA).
+
+
+## VideoLlavaConfig
+
+[[autodoc]] VideoLlavaConfig
+
+## VideoLlavaImageProcessor
+
+[[autodoc]] VideoLlavaImageProcessor
+
+## VideoLlavaProcessor
+
+[[autodoc]] VideoLlavaProcessor
+
+## VideoLlavaForConditionalGeneration
+
+[[autodoc]] VideoLlavaForConditionalGeneration
+    - forward

docs/source/en/perf_infer_gpu_one.md

@@ -55,6 +55,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [Llava](https://huggingface.co/docs/transformers/model_doc/llava)
 * [Llava-NeXT](https://huggingface.co/docs/transformers/model_doc/llava_next)
 * [VipLlava](https://huggingface.co/docs/transformers/model_doc/vipllava)
+* [VideoLlava](https://huggingface.co/docs/transformers/model_doc/video_llava)
 * [M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)
 * [MBart](https://huggingface.co/docs/transformers/model_doc/mbart#transformers.MBartModel)
 * [Mistral](https://huggingface.co/docs/transformers/model_doc/mistral#transformers.MistralModel)

src/transformers/__init__.py

@@ -726,6 +726,7 @@
         "UnivNetFeatureExtractor",
     ],
     "models.upernet": ["UperNetConfig"],
+    "models.video_llava": ["VideoLlavaConfig"],
     "models.videomae": ["VideoMAEConfig"],
     "models.vilt": [
         "ViltConfig",
@@ -3215,6 +3216,14 @@
             "UperNetPreTrainedModel",
         ]
     )
+    _import_structure["models.video_llava"].extend(
+        [
+            "VideoLlavaForConditionalGeneration",
+            "VideoLlavaImageProcessor",
+            "VideoLlavaPreTrainedModel",
+            "VideoLlavaProcessor",
+        ]
+    )
     _import_structure["models.videomae"].extend(
         [
             "VideoMAEForPreTraining",
@@ -5281,6 +5290,7 @@
         UnivNetFeatureExtractor,
     )
     from .models.upernet import UperNetConfig
+    from .models.video_llava import VideoLlavaConfig
     from .models.videomae import VideoMAEConfig
     from .models.vilt import (
         ViltConfig,
@@ -7384,6 +7394,12 @@
             UperNetForSemanticSegmentation,
             UperNetPreTrainedModel,
         )
+        from .models.video_llava import (
+            VideoLlavaForConditionalGeneration,
+            VideoLlavaImageProcessor,
+            VideoLlavaPreTrainedModel,
+            VideoLlavaProcessor,
+        )
         from .models.videomae import (
             VideoMAEForPreTraining,
             VideoMAEForVideoClassification,

src/transformers/image_utils.py

@@ -65,6 +65,9 @@
 ]  # noqa
 
 
+VideoInput = Union[np.ndarray, "torch.Tensor", List[np.ndarray], List["torch.Tensor"]]  # noqa
+
+
 class ChannelDimension(ExplicitEnum):
     FIRST = "channels_first"
     LAST = "channels_last"

src/transformers/models/__init__.py

@@ -240,6 +240,7 @@
     unispeech_sat,
     univnet,
     upernet,
+    video_llava,
     videomae,
     vilt,
     vipllava,

src/transformers/models/auto/configuration_auto.py

@@ -253,6 +253,7 @@
         ("univnet", "UnivNetConfig"),
         ("upernet", "UperNetConfig"),
         ("van", "VanConfig"),
+        ("video_llava", "VideoLlavaConfig"),
         ("videomae", "VideoMAEConfig"),
         ("vilt", "ViltConfig"),
         ("vipllava", "VipLlavaConfig"),
@@ -538,6 +539,7 @@
         ("univnet", "UnivNet"),
         ("upernet", "UPerNet"),
         ("van", "VAN"),
+        ("video_llava", "VideoLlava"),
         ("videomae", "VideoMAE"),
         ("vilt", "ViLT"),
         ("vipllava", "VipLlava"),

src/transformers/models/auto/image_processing_auto.py

@@ -115,6 +115,7 @@
         ("udop", "LayoutLMv3ImageProcessor"),
         ("upernet", "SegformerImageProcessor"),
         ("van", "ConvNextImageProcessor"),
+        ("video_llava", "VideoLlavaImageProcessor"),
         ("videomae", "VideoMAEImageProcessor"),
         ("vilt", "ViltImageProcessor"),
         ("vipllava", "CLIPImageProcessor"),

src/transformers/models/auto/modeling_auto.py

@@ -326,6 +326,7 @@
         ("tvlt", "TvltForPreTraining"),
         ("unispeech", "UniSpeechForPreTraining"),
         ("unispeech-sat", "UniSpeechSatForPreTraining"),
+        ("video_llava", "VideoLlavaForConditionalGeneration"),
         ("videomae", "VideoMAEForPreTraining"),
         ("vipllava", "VipLlavaForConditionalGeneration"),
         ("visual_bert", "VisualBertForPreTraining"),
@@ -696,6 +697,7 @@
         ("llava", "LlavaForConditionalGeneration"),
         ("llava_next", "LlavaNextForConditionalGeneration"),
         ("pix2struct", "Pix2StructForConditionalGeneration"),
+        ("video_llava", "VideoLlavaForConditionalGeneration"),
         ("vipllava", "VipLlavaForConditionalGeneration"),
         ("vision-encoder-decoder", "VisionEncoderDecoderModel"),
     ]

src/transformers/models/auto/processing_auto.py

@@ -89,6 +89,7 @@
         ("tvp", "TvpProcessor"),
         ("unispeech", "Wav2Vec2Processor"),
         ("unispeech-sat", "Wav2Vec2Processor"),
+        ("video_llava", "VideoLlavaProcessor"),
         ("vilt", "ViltProcessor"),
         ("vipllava", "LlavaProcessor"),
         ("vision-text-dual-encoder", "VisionTextDualEncoderProcessor"),

src/transformers/models/auto/tokenization_auto.py

@@ -462,6 +462,7 @@
                     "T5TokenizerFast" if is_tokenizers_available() else None,
                 ),
             ),
+            ("video_llava", ("LlamaTokenizer", "LlamaTokenizerFast" if is_tokenizers_available() else None)),
             ("vilt", ("BertTokenizer", "BertTokenizerFast" if is_tokenizers_available() else None)),
             ("vipllava", ("LlamaTokenizer", "LlamaTokenizerFast" if is_tokenizers_available() else None)),
             ("visual_bert", ("BertTokenizer", "BertTokenizerFast" if is_tokenizers_available() else None)),

src/transformers/models/video_llava/__init__.py

@@ -0,0 +1,71 @@
+# 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_video_llava": ["VideoLlavaConfig"],
+    "processing_video_llava": ["VideoLlavaProcessor"],
+}
+
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["image_processing_video_llava"] = ["VideoLlavaImageProcessor"]
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_video_llava"] = [
+        "VideoLlavaPreTrainedModel",
+        "VideoLlavaForConditionalGeneration",
+    ]
+
+if TYPE_CHECKING:
+    from .configuration_video_llava import (
+        VideoLlavaConfig,
+    )
+    from .image_processing_video_llava import VideoLlavaProcessor
+
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .image_processing_video_llava import VideoLlavaImageProcessor
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_video_llava import (
+            VideoLlavaForConditionalGeneration,
+            VideoLlavaPreTrainedModel,
+        )
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)