<!--Copyright 2022 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.
-->
# BLOOM
## Overview
The BLOOM model has been proposed with its various versions through the [BigScience Workshop](https://bigscience.huggingface.co/). BigScience is inspired by other open science initiatives where researchers have pooled their time and resources to collectively achieve a higher impact.
The architecture of BLOOM is essentially similar to GPT3 (auto-regressive model for next token prediction), but has been trained on 46 different languages and 13 programming languages.
Several smaller versions of the models have been trained on the same dataset. BLOOM is available in the following versions:
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with BLOOM. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<PipelineTag pipeline="text-generation"/>
- [`BloomForCausalLM`] is supported by this [causal language modeling example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#gpt-2gpt-and-causal-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
See also:
- [Causal language modeling task guide](../tasks/language_modeling)
<!--Copyright 2020 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.
-->
# BORT
## Overview
The BORT model was proposed in [Optimal Subarchitecture Extraction for BERT](https://arxiv.org/abs/2010.10499) by
Adrian de Wynter and Daniel J. Perry. It is an optimal subset of architectural parameters for the BERT, which the
authors refer to as "Bort".
The abstract from the paper is the following:
*We extract an optimal subset of architectural parameters for the BERT architecture from Devlin et al. (2018) by
applying recent breakthroughs in algorithms for neural architecture search. This optimal subset, which we refer to as
"Bort", is demonstrably smaller, having an effective (that is, not counting the embedding layer) size of 5.5% the
original BERT-large architecture, and 16% of the net size. Bort is also able to be pretrained in 288 GPU hours, which
is 1.2% of the time required to pretrain the highest-performing BERT parametric architectural variant, RoBERTa-large
(Liu et al., 2019), and about 33% of that of the world-record, in GPU hours, required to train BERT-large on the same
hardware. It is also 7.9x faster on a CPU, as well as being better performing than other compressed variants of the
architecture, and some of the non-compressed variants: it obtains performance improvements of between 0.3% and 31%,
absolute, with respect to BERT-large, on multiple public natural language understanding (NLU) benchmarks.*
Tips:
- BORT's model architecture is based on BERT, so one can refer to [BERT's documentation page](bert) for the
model's API as well as usage examples.
- BORT uses the RoBERTa tokenizer instead of the BERT tokenizer, so one can refer to [RoBERTa's documentation page](roberta) for the tokenizer's API as well as usage examples.
- BORT requires a specific fine-tuning algorithm, called [Agora](https://adewynter.github.io/notes/bort_algorithms_and_applications.html#fine-tuning-with-algebraic-topology) ,
that is sadly not open-sourced yet. It would be very useful for the community, if someone tries to implement the
algorithm to make BORT fine-tuning work.
This model was contributed by [stefan-it](https://huggingface.co/stefan-it). The original code can be found [here](https://github.com/alexa/bort/).
<!--Copyright 2023 The Intel Labs Team Authors, The Microsoft Research Team Authors and HuggingFace Inc. 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.
-->
# BridgeTower
## Overview
The BridgeTower model was proposed in [BridgeTower: Building Bridges Between Encoders in Vision-Language Representative Learning](https://arxiv.org/abs/2206.08657) by Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan. The goal of this model is to build a
bridge between each uni-modal encoder and the cross-modal encoder to enable comprehensive and detailed interaction at each layer of the cross-modal encoder thus achieving remarkable performance on various downstream tasks with almost negligible additional performance and computational costs.
This paper has been accepted to the [AAAI'23](https://aaai.org/Conferences/AAAI-23/) conference.
The abstract from the paper is the following:
*Vision-Language (VL) models with the TWO-TOWER architecture have dominated visual-language representation learning in recent years.
Current VL models either use lightweight uni-modal encoders and learn to extract, align and fuse both modalities simultaneously in a deep cross-modal encoder, or feed the last-layer uni-modal representations from the deep pre-trained uni-modal encoders into the top cross-modal encoder.
Both approaches potentially restrict vision-language representation learning and limit model performance. In this paper, we propose BRIDGETOWER, which introduces multiple bridge layers that build a connection between the top layers of uni-modal encoders and each layer of the crossmodal encoder.
This enables effective bottom-up cross-modal alignment and fusion between visual and textual representations of different semantic levels of pre-trained uni-modal encoders in the cross-modal encoder. Pre-trained with only 4M images, BRIDGETOWER achieves state-of-the-art performance on various downstream vision-language tasks.
In particular, on the VQAv2 test-std set, BRIDGETOWER achieves an accuracy of 78.73%, outperforming the previous state-of-the-art model METER by 1.09% with the same pre-training data and almost negligible additional parameters and computational costs.
Notably, when further scaling the model, BRIDGETOWER achieves an accuracy of 81.15%, surpassing models that are pre-trained on orders-of-magnitude larger datasets.*
<small> BridgeTower architecture. Taken from the <a href="https://arxiv.org/abs/2206.08657">original paper.</a> </small>
## Usage
BridgeTower consists of a visual encoder, a textual encoder and cross-modal encoder with multiple lightweight bridge layers.
The goal of this approach was to build a bridge between each uni-modal encoder and the cross-modal encoder to enable comprehensive and detailed interaction at each layer of the cross-modal encoder.
In principle, one can apply any visual, textual or cross-modal encoder in the proposed architecture.
The [`BridgeTowerProcessor`] wraps [`RobertaTokenizer`] and [`BridgeTowerImageProcessor`] into a single instance to both
encode the text and prepare the images respectively.
The following example shows how to run contrastive learning using [`BridgeTowerProcessor`] and [`BridgeTowerForContrastiveLearning`].
```python
>>> from transformers import BridgeTowerProcessor, BridgeTowerForContrastiveLearning
This model was contributed by [Anahita Bhiwandiwalla](https://huggingface.co/anahita-b), [Tiep Le](https://huggingface.co/Tile) and [Shaoyen Tseng](https://huggingface.co/shaoyent). The original code can be found [here](https://github.com/microsoft/BridgeTower).
Tips:
- This implementation of BridgeTower uses [`RobertaTokenizer`] to generate text embeddings and OpenAI's CLIP/ViT model to compute visual embeddings.
- Checkpoints for pre-trained [bridgeTower-base](https://huggingface.co/BridgeTower/bridgetower-base) and [bridgetower masked language modeling and image text matching](https://huggingface.co/BridgeTower/bridgetower-base-itm-mlm) are released.
- Please refer to [Table 5](https://arxiv.org/pdf/2206.08657.pdf) for BridgeTower's performance on Image Retrieval and other down stream tasks.
- The PyTorch version of this model is only available in torch 1.10 and higher.
<!--Copyright 2021 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.
-->
# ByT5
## Overview
The ByT5 model was presented in [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir
Kale, Adam Roberts, Colin Raffel.
The abstract from the paper is the following:
*Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units.
Encoding text as a sequence of tokens requires a tokenizer, which is typically created as an independent artifact from
the model. Token-free models that instead operate directly on raw text (bytes or characters) have many benefits: they
can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by
removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token
sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of
operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with
minimal modifications to process byte sequences. We carefully characterize the trade-offs in terms of parameter count,
training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level
counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on
tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of
pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our
experiments.*
This model was contributed by [patrickvonplaten](https://huggingface.co/patrickvonplaten). The original code can be
found [here](https://github.com/google-research/byt5).
ByT5's architecture is based on the T5v1.1 model, so one can refer to [T5v1.1's documentation page](t5v1.1). They
only differ in how inputs should be prepared for the model, see the code examples below.
Since ByT5 was pre-trained unsupervisedly, there's no real advantage to using a task prefix during single-task
fine-tuning. If you are doing multi-task fine-tuning, you should use a prefix.
### Example
ByT5 works on raw UTF-8 bytes, so it can be used without a tokenizer:
```python
>>> from transformers import T5ForConditionalGeneration
>>> import torch
>>> model = T5ForConditionalGeneration.from_pretrained("google/byt5-small")
>>> num_special_tokens = 3
>>> # Model has 3 special tokens which take up the input ids 0,1,2 of ByT5.
>>> # => Need to shift utf-8 character encodings by 3 before passing ids to model.
>>> input_ids = torch.tensor([list("Life is like a box of chocolates.".encode("utf-8"))]) + num_special_tokens
>>> labels = torch.tensor([list("La vie est comme une boîte de chocolat.".encode("utf-8"))]) + num_special_tokens
>>> loss = model(input_ids, labels=labels).loss
>>> loss.item()
2.66
```
For batched inference and training it is however recommended to make use of the tokenizer:
```python
>>> from transformers import T5ForConditionalGeneration, AutoTokenizer
>>> model = T5ForConditionalGeneration.from_pretrained("google/byt5-small")
This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found [here](https://github.com/google-research/language/tree/master/language/canine).
### Example
CANINE works on raw characters, so it can be used without a tokenizer:
```python
>>> from transformers import CanineModel
>>> import torch
>>> model = CanineModel.from_pretrained("google/canine-c") # model pre-trained with autoregressive character loss
>>> text = "hello world"
>>> # use Python's built-in ord() function to turn each character into its unicode code point id
>>> input_ids = torch.tensor([[ord(char) for char in text]])
>>> outputs = model(input_ids) # forward pass
>>> pooled_output = outputs.pooler_output
>>> sequence_output = outputs.last_hidden_state
```
For batched inference and training, it is however recommended to make use of the tokenizer (to pad/truncate all
sequences to the same length):
```python
>>> from transformers import CanineTokenizer, CanineModel
>>> model = CanineModel.from_pretrained("google/canine-c")
<!--Copyright 2022 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.
-->
# Chinese-CLIP
## Overview
The Chinese-CLIP model was proposed in [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
Chinese-CLIP is an implementation of CLIP (Radford et al., 2021) on a large-scale dataset of Chinese image-text pairs. It is capable of performing cross-modal retrieval and also playing as a vision backbone for vision tasks like zero-shot image classification, open-domain object detection, etc. The original Chinese-CLIP code is released [at this link](https://github.com/OFA-Sys/Chinese-CLIP).
The abstract from the paper is the following:
*The tremendous success of CLIP (Radford et al., 2021) has promoted the research and application of contrastive learning for vision-language pretraining. In this work, we construct a large-scale dataset of image-text pairs in Chinese, where most data are retrieved from publicly available datasets, and we pretrain Chinese CLIP models on the new dataset. We develop 5 Chinese CLIP models of multiple sizes, spanning from 77 to 958 million parameters. Furthermore, we propose a two-stage pretraining method, where the model is first trained with the image encoder frozen and then trained with all parameters being optimized, to achieve enhanced model performance. Our comprehensive experiments demonstrate that Chinese CLIP can achieve the state-of-the-art performance on MUGE, Flickr30K-CN, and COCO-CN in the setups of zero-shot learning and finetuning, and it is able to achieve competitive performance in zero-shot image classification based on the evaluation on the ELEVATER benchmark (Li et al., 2022). Our codes, pretrained models, and demos have been released.*
## Usage
The code snippet below shows how to compute image & text features and similarities:
```python
>>> from PIL import Image
>>> import requests
>>> from transformers import ChineseCLIPProcessor, ChineseCLIPModel
>>> model = ChineseCLIPModel.from_pretrained("OFA-Sys/chinese-clip-vit-base-patch16")
<!--Copyright 2023 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.
-->
# CLAP
## Overview
The CLAP model was proposed in [Large Scale Constrastive Laungaue-Audio pretraining with
feature fusion and keyword-to-caption augmentation](https://arxiv.org/pdf/2211.06687.pdf) by Yusong Wu, Ke Chen, Tianyu Zhang, Yuchen Hui, Taylor Berg-Kirkpatrick, Shlomo Dubnov.
CLAP (Constrastive Laungaue-Audio Pretraining) is a neural network trained on a variety of (audio, text) pairs. It can be instructed in to predict the most relevant text snippet, given an audio, without directly optimizing for the task. The CLAP model uses a SWINTransformer to get audio features from a log-Mel spectrogram input, and a RoBERTa model to get text features. Both the text and audio features are then projected to a latent space with identical dimension. The dot product between the projected audio and text features is then used as a similar score.
The abstract from the paper is the following:
*Contrastive learning has shown remarkable success in the field of multimodal representation learning. In this paper, we propose a pipeline of contrastive language-audio pretraining to develop an audio representation by combining audio data with natural language descriptions. To accomplish this target, we first release LAION-Audio-630K, a large collection of 633,526 audio-text pairs from different data sources. Second, we construct a contrastive language-audio pretraining model by considering different audio encoders and text encoders. We incorporate the feature fusion mechanism and keyword-to-caption augmentation into the model design to further enable the model to process audio inputs of variable lengths and enhance the performance. Third, we perform comprehensive experiments to evaluate our model across three tasks: text-to-audio retrieval, zero-shot audio classification, and supervised audio classification. The results demonstrate that our model achieves superior performance in text-to-audio retrieval task. In audio classification tasks, the model achieves state-of-the-art performance in the zeroshot setting and is able to obtain performance comparable to models' results in the non-zero-shot setting. LAION-Audio-6*
This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArtZucker) .
The original code can be found [here](https://github.com/LAION-AI/Clap).
AlistofofficialHuggingFaceandcommunity(indicatedby🌎)resourcestohelpyougetstartedwithCLIPSeg.Ifyou're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'llreviewit! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<!--Copyright 2022 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.
-->
# CodeGen
## Overview
The CodeGen model was proposed in [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, and Caiming Xiong.
CodeGen is an autoregressive language model for program synthesis trained sequentially on [The Pile](https://pile.eleuther.ai/), BigQuery, and BigPython.
The abstract from the paper is the following:
*Program synthesis strives to generate a computer program as a solution to a given problem specification. We propose a conversational program synthesis approach via large language models, which addresses the challenges of searching over a vast program space and user intent specification faced in prior approaches. Our new approach casts the process of writing a specification and program as a multi-turn conversation between a user and a system. It treats program synthesis as a sequence prediction problem, in which the specification is expressed in natural language and the desired program is conditionally sampled. We train a family of large language models, called CodeGen, on natural language and programming language data. With weak supervision in the data and the scaling up of data size and model size, conversational capacities emerge from the simple autoregressive language modeling. To study the model behavior on conversational program synthesis, we develop a multi-turn programming benchmark (MTPB), where solving each problem requires multi-step synthesis via multi-turn conversation between the user and the model. Our findings show the emergence of conversational capabilities and the effectiveness of the proposed conversational program synthesis paradigm. In addition, our model CodeGen (with up to 16B parameters trained on TPU-v4) outperforms OpenAI's Codex on the HumanEval benchmark. We make the training library JaxFormer including checkpoints available as open source contribution: [this https URL](https://github.com/salesforce/codegen).*
This model was contributed by [Hiroaki Hayashi](https://huggingface.co/rooa).
The original code can be found [here](https://github.com/salesforce/codegen).
## Checkpoint Naming
* CodeGen model [checkpoints](https://huggingface.co/models?other=codegen) are available on different pre-training data with variable sizes.
* The format is: `Salesforce/codegen-{size}-{data}`, where
* `size`: `350M`, `2B`, `6B`, `16B`
* `data`:
* `nl`: Pre-trained on the Pile
* `multi`: Initialized with `nl`, then further pre-trained on multiple programming languages data
* `mono`: Initialized with `multi`, then further pre-trained on Python data
* For example, `Salesforce/codegen-350M-mono` offers a 350 million-parameter checkpoint pre-trained sequentially on the Pile, multiple programming languages, and Python.
## How to use
```python
>>> from transformers import AutoModelForCausalLM, AutoTokenizer
>>> checkpoint = "Salesforce/codegen-350M-mono"
>>> model = AutoModelForCausalLM.from_pretrained(checkpoint)
<!--Copyright 2022 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.
-->
# Conditional DETR
## Overview
The Conditional DETR model was proposed in [Conditional DETR for Fast Training Convergence](https://arxiv.org/abs/2108.06152) by Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, Jingdong Wang. Conditional DETR presents a conditional cross-attention mechanism for fast DETR training. Conditional DETR converges 6.7× to 10× faster than DETR.
The abstract from the paper is the following:
*The recently-developed DETR approach applies the transformer encoder and decoder architecture to object detection and achieves promising performance. In this paper, we handle the critical issue, slow training convergence, and present a conditional cross-attention mechanism for fast DETR training. Our approach is motivated by that the cross-attention in DETR relies highly on the content embeddings for localizing the four extremities and predicting the box, which increases the need for high-quality content embeddings and thus the training difficulty. Our approach, named conditional DETR, learns a conditional spatial query from the decoder embedding for decoder multi-head cross-attention. The benefit is that through the conditional spatial query, each cross-attention head is able to attend to a band containing a distinct region, e.g., one object extremity or a region inside the object box. This narrows down the spatial range for localizing the distinct regions for object classification and box regression, thus relaxing the dependence on the content embeddings and easing the training. Empirical results show that conditional DETR converges 6.7× faster for the backbones R50 and R101 and 10× faster for stronger backbones DC5-R50 and DC5-R101. Code is available at https://github.com/Atten4Vis/ConditionalDETR.*
<small> Conditional DETR shows much faster convergence compared to the original DETR. Taken from the <a href="https://arxiv.org/abs/2108.06152">original paper</a>.</small>
This model was contributed by [DepuMeng](https://huggingface.co/DepuMeng). The original code can be found [here](https://github.com/Atten4Vis/ConditionalDETR).
The ConvBERT model was proposed in [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng
Yan.
The abstract from the paper is the following:
*Pre-trained language models like BERT and its variants have recently achieved impressive performance in various
natural language understanding tasks. However, BERT heavily relies on the global self-attention block and thus suffers
large memory footprint and computation cost. Although all its attention heads query on the whole input sequence for
generating the attention map from a global perspective, we observe some heads only need to learn local dependencies,
which means the existence of computation redundancy. We therefore propose a novel span-based dynamic convolution to
replace these self-attention heads to directly model local dependencies. The novel convolution heads, together with the
rest self-attention heads, form a new mixed attention block that is more efficient at both global and local context
learning. We equip BERT with this mixed attention design and build a ConvBERT model. Experiments have shown that
ConvBERT significantly outperforms BERT and its variants in various downstream tasks, with lower training cost and
fewer model parameters. Remarkably, ConvBERTbase model achieves 86.4 GLUE score, 0.7 higher than ELECTRAbase, while
using less than 1/4 training cost. Code and pre-trained models will be released.*
ConvBERT training tips are similar to those of BERT.
This model was contributed by [abhishek](https://huggingface.co/abhishek). The original implementation can be found
<!--Copyright 2022 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.
-->
# ConvNeXT
## Overview
The ConvNeXT model was proposed in [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
ConvNeXT is a pure convolutional model (ConvNet), inspired by the design of Vision Transformers, that claims to outperform them.
The abstract from the paper is the following:
*The "Roaring 20s" of visual recognition began with the introduction of Vision Transformers (ViTs), which quickly superseded ConvNets as the state-of-the-art image classification model.
A vanilla ViT, on the other hand, faces difficulties when applied to general computer vision tasks such as object detection and semantic segmentation. It is the hierarchical Transformers
(e.g., Swin Transformers) that reintroduced several ConvNet priors, making Transformers practically viable as a generic vision backbone and demonstrating remarkable performance on a wide
variety of vision tasks. However, the effectiveness of such hybrid approaches is still largely credited to the intrinsic superiority of Transformers, rather than the inherent inductive
biases of convolutions. In this work, we reexamine the design spaces and test the limits of what a pure ConvNet can achieve. We gradually "modernize" a standard ResNet toward the design
of a vision Transformer, and discover several key components that contribute to the performance difference along the way. The outcome of this exploration is a family of pure ConvNet models
dubbed ConvNeXt. Constructed entirely from standard ConvNet modules, ConvNeXts compete favorably with Transformers in terms of accuracy and scalability, achieving 87.8% ImageNet top-1 accuracy
and outperforming Swin Transformers on COCO detection and ADE20K segmentation, while maintaining the simplicity and efficiency of standard ConvNets.*
Tips:
- See the code examples below each model regarding usage.
<small> ConvNeXT architecture. Taken from the <a href="https://arxiv.org/abs/2201.03545">original paper</a>.</small>
This model was contributed by [nielsr](https://huggingface.co/nielsr). TensorFlow version of the model was contributed by [ariG23498](https://github.com/ariG23498),
[gante](https://github.com/gante), and [sayakpaul](https://github.com/sayakpaul) (equal contribution). The original code can be found [here](https://github.com/facebookresearch/ConvNeXt).
## Resources
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with ConvNeXT.
<PipelineTag pipeline="image-classification"/>
- [`ConvNextForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
- See also: [Image classification task guide](../tasks/image_classification)
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<!--Copyright 2023 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.
-->
# ConvNeXt V2
## Overview
The ConvNeXt V2 model was proposed in [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) by Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, Saining Xie.
ConvNeXt V2 is a pure convolutional model (ConvNet), inspired by the design of Vision Transformers, and a successor of [ConvNeXT](convnext).
The abstract from the paper is the following:
*Driven by improved architectures and better representation learning frameworks, the field of visual recognition has enjoyed rapid modernization and performance boost in the early 2020s. For example, modern ConvNets, represented by ConvNeXt, have demonstrated strong performance in various scenarios. While these models were originally designed for supervised learning with ImageNet labels, they can also potentially benefit from self-supervised learning techniques such as masked autoencoders (MAE). However, we found that simply combining these two approaches leads to subpar performance. In this paper, we propose a fully convolutional masked autoencoder framework and a new Global Response Normalization (GRN) layer that can be added to the ConvNeXt architecture to enhance inter-channel feature competition. This co-design of self-supervised learning techniques and architectural improvement results in a new model family called ConvNeXt V2, which significantly improves the performance of pure ConvNets on various recognition benchmarks, including ImageNet classification, COCO detection, and ADE20K segmentation. We also provide pre-trained ConvNeXt V2 models of various sizes, ranging from an efficient 3.7M-parameter Atto model with 76.7% top-1 accuracy on ImageNet, to a 650M Huge model that achieves a state-of-the-art 88.9% accuracy using only public training data.*
Tips:
- See the code examples below each model regarding usage.
<small> ConvNeXt V2 architecture. Taken from the <a href="https://arxiv.org/abs/2301.00808">original paper</a>.</small>
This model was contributed by [adirik](https://huggingface.co/adirik). The original code can be found [here](https://github.com/facebookresearch/ConvNeXt-V2).
## Resources
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with ConvNeXt V2.
<PipelineTag pipeline="image-classification"/>
- [`ConvNextV2ForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<!--Copyright 2020 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.
-->
# CPM
## Overview
The CPM model was proposed in [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin,
Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen,
CTRL model was proposed in [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and
Richard Socher. It's a causal (unidirectional) transformer pre-trained using language modeling on a very large corpus
of ~140 GB of text data with the first token reserved as a control code (such as Links, Books, Wikipedia etc.).
The abstract from the paper is the following:
*Large-scale language models show promising text generation capabilities, but users cannot easily control particular
aspects of the generated text. We release CTRL, a 1.63 billion-parameter conditional transformer language model,
trained to condition on control codes that govern style, content, and task-specific behavior. Control codes were
derived from structure that naturally co-occurs with raw text, preserving the advantages of unsupervised learning while
providing more explicit control over text generation. These codes also allow CTRL to predict which parts of the
training data are most likely given a sequence. This provides a potential method for analyzing large amounts of data
via model-based source attribution.*
Tips:
- CTRL makes use of control codes to generate text: it requires generations to be started by certain words, sentences
or links to generate coherent text. Refer to the [original implementation](https://github.com/salesforce/ctrl) for
more information.
- CTRL is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
the left.
- CTRL was trained with a causal language modeling (CLM) objective and is therefore powerful at predicting the next
token in a sequence. Leveraging this feature allows CTRL to generate syntactically coherent text as it can be
observed in the *run_generation.py* example script.
- The PyTorch models can take the `past_key_values` as input, which is the previously computed key/value attention pairs.
TensorFlow models accepts `past` as input. Using the `past_key_values` value prevents the model from re-computing
pre-computed values in the context of text generation. See the [`forward`](model_doc/ctrl#transformers.CTRLModel.forward)
method for more information on the usage of this argument.
This model was contributed by [keskarnitishr](https://huggingface.co/keskarnitishr). The original code can be found
<!--Copyright 2022 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.
-->
# Convolutional Vision Transformer (CvT)
## Overview
The CvT model was proposed in [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan and Lei Zhang. The Convolutional vision Transformer (CvT) improves the [Vision Transformer (ViT)](vit) in performance and efficiency by introducing convolutions into ViT to yield the best of both designs.
The abstract from the paper is the following:
*We present in this paper a new architecture, named Convolutional vision Transformer (CvT), that improves Vision Transformer (ViT)
in performance and efficiency by introducing convolutions into ViT to yield the best of both designs. This is accomplished through
two primary modifications: a hierarchy of Transformers containing a new convolutional token embedding, and a convolutional Transformer
block leveraging a convolutional projection. These changes introduce desirable properties of convolutional neural networks (CNNs)
to the ViT architecture (\ie shift, scale, and distortion invariance) while maintaining the merits of Transformers (\ie dynamic attention,
global context, and better generalization). We validate CvT by conducting extensive experiments, showing that this approach achieves
state-of-the-art performance over other Vision Transformers and ResNets on ImageNet-1k, with fewer parameters and lower FLOPs. In addition,
performance gains are maintained when pretrained on larger datasets (\eg ImageNet-22k) and fine-tuned to downstream tasks. Pre-trained on
ImageNet-22k, our CvT-W24 obtains a top-1 accuracy of 87.7\% on the ImageNet-1k val set. Finally, our results show that the positional encoding,
a crucial component in existing Vision Transformers, can be safely removed in our model, simplifying the design for higher resolution vision tasks.*
Tips:
- CvT models are regular Vision Transformers, but trained with convolutions. They outperform the [original model (ViT)](vit) when fine-tuned on ImageNet-1K and CIFAR-100.
- You can check out demo notebooks regarding inference as well as fine-tuning on custom data [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/VisionTransformer) (you can just replace [`ViTFeatureExtractor`] by [`AutoImageProcessor`] and [`ViTForImageClassification`] by [`CvtForImageClassification`]).
- The available checkpoints are either (1) pre-trained on [ImageNet-22k](http://www.image-net.org/) (a collection of 14 million images and 22k classes) only, (2) also fine-tuned on ImageNet-22k or (3) also fine-tuned on [ImageNet-1k](http://www.image-net.org/challenges/LSVRC/2012/) (also referred to as ILSVRC 2012, a collection of 1.3 million
images and 1,000 classes).
This model was contributed by [anugunj](https://huggingface.co/anugunj). The original code can be found [here](https://github.com/microsoft/CvT).
## Resources
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with CvT.
<PipelineTag pipeline="image-classification"/>
- [`CvtForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
- See also: [Image classification task guide](../tasks/image_classification)
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<!--Copyright 2022 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.
-->
# Data2Vec
## Overview
The Data2Vec model was proposed in [data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/pdf/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu and Michael Auli.
Data2Vec proposes a unified framework for self-supervised learning across different data modalities - text, audio and images.
Importantly, predicted targets for pre-training are contextualized latent representations of the inputs, rather than modality-specific, context-independent targets.
The abstract from the paper is the following:
*While the general idea of self-supervised learning is identical across modalities, the actual algorithms and
objectives differ widely because they were developed with a single modality in mind. To get us closer to general
self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech,
NLP or computer vision. The core idea is to predict latent representations of the full input data based on a
masked view of the input in a selfdistillation setup using a standard Transformer architecture.
Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which
are local in nature, data2vec predicts contextualized latent representations that contain information from
the entire input. Experiments on the major benchmarks of speech recognition, image classification, and
natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches.
Models and code are available at www.github.com/pytorch/fairseq/tree/master/examples/data2vec.*
Tips:
- Data2VecAudio, Data2VecText, and Data2VecVision have all been trained using the same self-supervised learning method.
- For Data2VecAudio, preprocessing is identical to [`Wav2Vec2Model`], including feature extraction
- For Data2VecText, preprocessing is identical to [`RobertaModel`], including tokenization.
- For Data2VecVision, preprocessing is identical to [`BeitModel`], including feature extraction.
This model was contributed by [edugp](https://huggingface.co/edugp) and [patrickvonplaten](https://huggingface.co/patrickvonplaten).
[sayakpaul](https://github.com/sayakpaul) and [Rocketknight1](https://github.com/Rocketknight1) contributed Data2Vec for vision in TensorFlow.
The original code (for NLP and Speech) can be found [here](https://github.com/pytorch/fairseq/tree/main/examples/data2vec).
The original code for vision can be found [here](https://github.com/facebookresearch/data2vec_vision/tree/main/beit).
## Resources
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Data2Vec.
<PipelineTag pipeline="image-classification"/>
- [`Data2VecVisionForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
- To fine-tune [`TFData2VecVisionForImageClassification`] on a custom dataset, see [this notebook](https://colab.research.google.com/github/sayakpaul/TF-2.0-Hacks/blob/master/data2vec_vision_image_classification.ipynb).
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<!--Copyright 2020 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.
-->
# DeBERTa-v2
## Overview
The DeBERTa model was proposed in [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen It is based on Google's
BERT model released in 2018 and Facebook's RoBERTa model released in 2019.
It builds on RoBERTa with disentangled attention and enhanced mask decoder training with half of the data used in
RoBERTa.
The abstract from the paper is the following:
*Recent progress in pre-trained neural language models has significantly improved the performance of many natural
language processing (NLP) tasks. In this paper we propose a new model architecture DeBERTa (Decoding-enhanced BERT with
disentangled attention) that improves the BERT and RoBERTa models using two novel techniques. The first is the
disentangled attention mechanism, where each word is represented using two vectors that encode its content and
position, respectively, and the attention weights among words are computed using disentangled matrices on their
contents and relative positions. Second, an enhanced mask decoder is used to replace the output softmax layer to
predict the masked tokens for model pretraining. We show that these two techniques significantly improve the efficiency
of model pretraining and performance of downstream tasks. Compared to RoBERTa-Large, a DeBERTa model trained on half of
the training data performs consistently better on a wide range of NLP tasks, achieving improvements on MNLI by +0.9%
(90.2% vs. 91.1%), on SQuAD v2.0 by +2.3% (88.4% vs. 90.7%) and RACE by +3.6% (83.2% vs. 86.8%). The DeBERTa code and
pre-trained models will be made publicly available at https://github.com/microsoft/DeBERTa.*
The following information is visible directly on the [original implementation
repository](https://github.com/microsoft/DeBERTa). DeBERTa v2 is the second version of the DeBERTa model. It includes
the 1.5B model used for the SuperGLUE single-model submission and achieving 89.9, versus human baseline 89.8. You can
find more details about this submission in the authors'
