Add MobileViTv2 (#22820)

* generated code from add-new-model-like * Add code for modeling, config, and weight conversion * add tests for image-classification, update modeling and config * add code, tests for semantic-segmentation * make style, make quality, make fix-copies * make fix-copies * Update modeling_mobilevitv2.py fix bugs * Update _toctree.yml * update modeling, config fix bugs * Edit docs - fix bug MobileViTv2v2 -> MobileViTv2 * Update mobilevitv2.mdx * update docstrings * Update configuration_mobilevitv2.py make style * Update convert_mlcvnets_to_pytorch.py remove unused options * Update convert_mlcvnets_to_pytorch.py make style * Add suggestions from code review Co-Authored-By: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * make style, make quality * Add suggestions from code review Co-Authored-By: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Add suggestions from code review Remove MobileViTv2ImageProcessor Co-Authored-By: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * make style * Add suggestions from code review Rename MobileViTv2 -> MobileViTV2 Co-Authored-By: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Add suggestions from code review Co-Authored-By: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update modeling_mobilevitv2.py make style * Update serialization.mdx * Update modeling_mobilevitv2.py --------- Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
2025-07-31 02:02:21 +06:00 · 2023-06-02 15:07:02 +05:30 · 2023-06-02 15:07:02 +05:30 · 07c54413ac
commit 07c54413ac
parent 5dfd407b37
24 changed files with 2116 additions and 3 deletions
--- a/README.md
+++ b/README.md
@ -406,6 +406,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
--- a/README_es.md
+++ b/README_es.md
@ -381,6 +381,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
--- a/README_hd.md
+++ b/README_hd.md
@ -353,6 +353,7 @@ conda install -c huggingface transformers
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (Apple से) साथ में कागज [MobileViT: लाइट-वेट, जनरल-पर्पस, और मोबाइल-फ्रेंडली विजन ट्रांसफॉर्मर] (https://arxiv.org/abs/2110.02178) सचिन मेहता और मोहम्मद रस्तगरी द्वारा पोस्ट किया गया।
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple से) Sachin Mehta and Mohammad Rastegari. द्वाराअनुसंधान पत्र [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) के साथ जारी किया गया
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI से) साथ वाला पेपर [mT5: एक व्यापक बहुभाषी पूर्व-प्रशिक्षित टेक्स्ट-टू-टेक्स्ट ट्रांसफॉर्मर]( https://arxiv.org/abs/2010.11934) लिंटिंग ज़ू, नोआ कॉन्सटेंट, एडम रॉबर्ट्स, मिहिर काले, रामी अल-रफू, आदित्य सिद्धांत, आदित्य बरुआ, कॉलिन रैफेल द्वारा पोस्ट किया गया।
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
--- a/README_ja.md
+++ b/README_ja.md
@ -415,6 +415,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (Google Inc. から) Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam から公開された研究論文: [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861)
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (Google Inc. から) Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen から公開された研究論文: [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381)
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (Apple から) Sachin Mehta and Mohammad Rastegari から公開された研究論文: [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178)
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple から) Sachin Mehta and Mohammad Rastegari. から公開された研究論文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (Microsoft Research から) Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu から公開された研究論文: [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297)
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI から) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel から公開された研究論文: [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934)
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box から) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen から公開された研究論文: [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131)
--- a/README_ko.md
+++ b/README_ko.md
@ -330,6 +330,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (Google Inc. 에서) Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam 의 [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) 논문과 함께 발표했습니다.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (Google Inc. 에서) Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen 의 [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) 논문과 함께 발표했습니다.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (Apple 에서) Sachin Mehta and Mohammad Rastegari 의 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 논문과 함께 발표했습니다.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (Apple 에서 제공)은 Sachin Mehta and Mohammad Rastegari.의 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680)논문과 함께 발표했습니다.
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (Microsoft Research 에서) Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 의 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 논문과 함께 발표했습니다.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (Google AI 에서) Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 의 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 논문과 함께 발표했습니다.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (RUC AI Box 에서) Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 의 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 논문과 함께 발표했습니다.
--- a/README_zh-hans.md
+++ b/README_zh-hans.md
@ -340,7 +340,7 @@ conda install -c huggingface transformers
 1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** 用 [OPUS](http://opus.nlpl.eu/) 数据训练的机器翻译模型由 Jörg Tiedemann 发布。[Marian Framework](https://marian-nmt.github.io/) 由微软翻译团队开发。
 1. **[MarkupLM](https://huggingface.co/docs/transformers/model_doc/markuplm)** (来自 Microsoft Research Asia) 伴随论文 [MarkupLM: Pre-training of Text and Markup Language for Visually-rich Document Understanding](https://arxiv.org/abs/2110.08518) 由 Junlong Li, Yiheng Xu, Lei Cui, Furu Wei 发布。
 1. **[Mask2Former](https://huggingface.co/docs/transformers/model_doc/mask2former)** (来自 FAIR and UIUC) 伴随论文 [Masked-attention Mask Transformer for Universal Image Segmentation](https://arxiv.org/abs/2112.01527) 由 Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, Rohit Girdhar 发布。
-1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov  
 1. **[MatCha](https://huggingface.co/docs/transformers/model_doc/matcha)** (来自 Google AI) 伴随论文 [MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering](https://arxiv.org/abs/2212.09662) 由 Fangyu Liu, Francesco Piccinno, Syrine Krichene, Chenxi Pang, Kenton Lee, Mandar Joshi, Yasemin Altun, Nigel Collier, Julian Martin Eisenschlos 发布。
 1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) 由 Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer 发布。
 1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) 由 Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan 发布。
@ -354,6 +354,7 @@ conda install -c huggingface transformers
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (来自 Google Inc.) 伴随论文 [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) 由 Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam 发布。
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (来自 Google Inc.) 伴随论文 [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) 由 Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen 发布。
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (来自 Apple) 伴随论文 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 由 Sachin Mehta and Mohammad Rastegari 发布。
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (来自 Apple) 伴随论文 [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) 由 Sachin Mehta and Mohammad Rastegari 发布。
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (来自 Microsoft Research) 伴随论文 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 由 Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 发布。
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。
--- a/README_zh-hant.md
+++ b/README_zh-hant.md
@ -366,6 +366,7 @@ conda install -c huggingface transformers
 1. **[MobileNetV1](https://huggingface.co/docs/transformers/model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
 1. **[MobileNetV2](https://huggingface.co/docs/transformers/model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
 1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](https://huggingface.co/docs/transformers/main/model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
 1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
--- a/docs/source/en/_toctree.yml
+++ b/docs/source/en/_toctree.yml
@ -494,6 +494,8 @@
        title: MobileNetV2
      - local: model_doc/mobilevit
        title: MobileViT
+      - local: model_doc/mobilevitv2
+        title: MobileViTV2
      - local: model_doc/nat
        title: NAT
      - local: model_doc/poolformer
--- a/docs/source/en/index.mdx
+++ b/docs/source/en/index.mdx
@ -167,6 +167,7 @@ The documentation is organized into five sections:
 1. **[MobileNetV1](model_doc/mobilenet_v1)** (from Google Inc.) released with the paper [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://arxiv.org/abs/1704.04861) by Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam.
 1. **[MobileNetV2](model_doc/mobilenet_v2)** (from Google Inc.) released with the paper [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://arxiv.org/abs/1801.04381) by Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen.
 1. **[MobileViT](model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MobileViTV2](model_doc/mobilevitv2)** (from Apple) released with the paper [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
 1. **[MPNet](model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
 1. **[MT5](model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
 1. **[MVP](model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
@ -369,6 +370,7 @@ Flax), PyTorch, and/or TensorFlow.
 |          MobileNetV1          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |          MobileNetV2          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           MobileViT           |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
+|          MobileViTV2          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |             MPNet             |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |              MT5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |              MVP              |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
--- a/docs/source/en/model_doc/mobilevitv2.mdx
+++ b/docs/source/en/model_doc/mobilevitv2.mdx
@ -0,0 +1,53 @@
+<!--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.
+-->
+
+# MobileViTV2
+
+## Overview
+
+The MobileViTV2 model was proposed in [Separable Self-attention for Mobile Vision Transformers](https://arxiv.org/abs/2206.02680) by Sachin Mehta and Mohammad Rastegari.
+
+MobileViTV2 is the second version of MobileViT, constructed by replacing the multi-headed self-attention in MobileViT with separable self-attention.
+
+The abstract from the paper is the following:
+
+*Mobile vision transformers (MobileViT) can achieve state-of-the-art performance across several mobile vision tasks, including classification and detection. Though these models have fewer parameters, they have high latency as compared to convolutional neural network-based models. The main efficiency bottleneck in MobileViT is the multi-headed self-attention (MHA) in transformers, which requires O(k2) time complexity with respect to the number of tokens (or patches) k. Moreover, MHA requires costly operations (e.g., batch-wise matrix multiplication) for computing self-attention, impacting latency on resource-constrained devices. This paper introduces a separable self-attention method with linear complexity, i.e. O(k). A simple yet effective characteristic of the proposed method is that it uses element-wise operations for computing self-attention, making it a good choice for resource-constrained devices. The improved model, MobileViTV2, is state-of-the-art on several mobile vision tasks, including ImageNet object classification and MS-COCO object detection. With about three million parameters, MobileViTV2 achieves a top-1 accuracy of 75.6% on the ImageNet dataset, outperforming MobileViT by about 1% while running 3.2× faster on a mobile device.*
+
+Tips:
+
+- MobileViTV2 is more like a CNN than a Transformer model. It does not work on sequence data but on batches of images. Unlike ViT, there are no embeddings. The backbone model outputs a feature map.
+- One can use [`MobileViTImageProcessor`] to prepare images for the model. Note that if you do your own preprocessing, the pretrained checkpoints expect images to be in BGR pixel order (not RGB).
+- The available image classification checkpoints are pre-trained on [ImageNet-1k](https://huggingface.co/datasets/imagenet-1k) (also referred to as ILSVRC 2012, a collection of 1.3 million images and 1,000 classes).
+- The segmentation model uses a [DeepLabV3](https://arxiv.org/abs/1706.05587) head. The available semantic segmentation checkpoints are pre-trained on [PASCAL VOC](http://host.robots.ox.ac.uk/pascal/VOC/).
+
+This model was contributed by [shehan97](https://huggingface.co/shehan97).
+The original code can be found [here](https://github.com/apple/ml-cvnets).
+
+
+## MobileViTV2Config
+
+[[autodoc]] MobileViTV2Config
+
+## MobileViTV2Model
+
+[[autodoc]] MobileViTV2Model
+    - forward
+
+## MobileViTV2ForImageClassification
+
+[[autodoc]] MobileViTV2ForImageClassification
+    - forward
+
+## MobileViTV2ForSemanticSegmentation
+
+[[autodoc]] MobileViTV2ForSemanticSegmentation
+    - forward
--- a/docs/source/en/tasks/image_classification.mdx
+++ b/docs/source/en/tasks/image_classification.mdx
@ -30,7 +30,7 @@ The task illustrated in this tutorial is supported by the following model archit

 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->

-[BEiT](../model_doc/beit), [BiT](../model_doc/bit), [ConvNeXT](../model_doc/convnext), [ConvNeXTV2](../model_doc/convnextv2), [CvT](../model_doc/cvt), [Data2VecVision](../model_doc/data2vec-vision), [DeiT](../model_doc/deit), [DiNAT](../model_doc/dinat), [EfficientFormer](../model_doc/efficientformer), [EfficientNet](../model_doc/efficientnet), [FocalNet](../model_doc/focalnet), [ImageGPT](../model_doc/imagegpt), [LeViT](../model_doc/levit), [MobileNetV1](../model_doc/mobilenet_v1), [MobileNetV2](../model_doc/mobilenet_v2), [MobileViT](../model_doc/mobilevit), [NAT](../model_doc/nat), [Perceiver](../model_doc/perceiver), [PoolFormer](../model_doc/poolformer), [RegNet](../model_doc/regnet), [ResNet](../model_doc/resnet), [SegFormer](../model_doc/segformer), [SwiftFormer](../model_doc/swiftformer), [Swin Transformer](../model_doc/swin), [Swin Transformer V2](../model_doc/swinv2), [VAN](../model_doc/van), [ViT](../model_doc/vit), [ViT Hybrid](../model_doc/vit_hybrid), [ViTMSN](../model_doc/vit_msn)
+[BEiT](../model_doc/beit), [BiT](../model_doc/bit), [ConvNeXT](../model_doc/convnext), [ConvNeXTV2](../model_doc/convnextv2), [CvT](../model_doc/cvt), [Data2VecVision](../model_doc/data2vec-vision), [DeiT](../model_doc/deit), [DiNAT](../model_doc/dinat), [EfficientFormer](../model_doc/efficientformer), [EfficientNet](../model_doc/efficientnet), [FocalNet](../model_doc/focalnet), [ImageGPT](../model_doc/imagegpt), [LeViT](../model_doc/levit), [MobileNetV1](../model_doc/mobilenet_v1), [MobileNetV2](../model_doc/mobilenet_v2), [MobileViT](../model_doc/mobilevit), [MobileViTV2](../model_doc/mobilevitv2), [NAT](../model_doc/nat), [Perceiver](../model_doc/perceiver), [PoolFormer](../model_doc/poolformer), [RegNet](../model_doc/regnet), [ResNet](../model_doc/resnet), [SegFormer](../model_doc/segformer), [SwiftFormer](../model_doc/swiftformer), [Swin Transformer](../model_doc/swin), [Swin Transformer V2](../model_doc/swinv2), [VAN](../model_doc/van), [ViT](../model_doc/vit), [ViT Hybrid](../model_doc/vit_hybrid), [ViTMSN](../model_doc/vit_msn)
 <!--End of the generated tip-->

 </Tip>
--- a/docs/source/en/tasks/semantic_segmentation.mdx
+++ b/docs/source/en/tasks/semantic_segmentation.mdx
@ -28,7 +28,7 @@ The task illustrated in this tutorial is supported by the following model archit

 <!--This tip is automatically generated by `make fix-copies`, do not fill manually!-->

-[BEiT](../model_doc/beit), [Data2VecVision](../model_doc/data2vec-vision), [DPT](../model_doc/dpt), [MobileNetV2](../model_doc/mobilenet_v2), [MobileViT](../model_doc/mobilevit), [SegFormer](../model_doc/segformer), [UPerNet](../model_doc/upernet)
+[BEiT](../model_doc/beit), [Data2VecVision](../model_doc/data2vec-vision), [DPT](../model_doc/dpt), [MobileNetV2](../model_doc/mobilenet_v2), [MobileViT](../model_doc/mobilevit), [MobileViTV2](../model_doc/mobilevitv2), [SegFormer](../model_doc/segformer), [UPerNet](../model_doc/upernet)

 <!--End of the generated tip-->

--- a/src/transformers/init.py
+++ b/src/transformers/init.py
@ -387,6 +387,7 @@ _import_structure = {
    "models.mobilenet_v1": ["MOBILENET_V1_PRETRAINED_CONFIG_ARCHIVE_MAP", "MobileNetV1Config"],
    "models.mobilenet_v2": ["MOBILENET_V2_PRETRAINED_CONFIG_ARCHIVE_MAP", "MobileNetV2Config"],
    "models.mobilevit": ["MOBILEVIT_PRETRAINED_CONFIG_ARCHIVE_MAP", "MobileViTConfig"],
+    "models.mobilevitv2": ["MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP", "MobileViTV2Config"],
    "models.mpnet": ["MPNET_PRETRAINED_CONFIG_ARCHIVE_MAP", "MPNetConfig", "MPNetTokenizer"],
    "models.mt5": ["MT5Config"],
    "models.mvp": ["MvpConfig", "MvpTokenizer"],
@ -2073,6 +2074,15 @@ else:
            "MobileViTPreTrainedModel",
        ]
    )
+    _import_structure["models.mobilevitv2"].extend(
+        [
+            "MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "MobileViTV2ForImageClassification",
+            "MobileViTV2ForSemanticSegmentation",
+            "MobileViTV2Model",
+            "MobileViTV2PreTrainedModel",
+        ]
+    )
    _import_structure["models.mpnet"].extend(
        [
            "MPNET_PRETRAINED_MODEL_ARCHIVE_LIST",
@ -4183,6 +4193,7 @@ if TYPE_CHECKING:
    from .models.mobilenet_v1 import MOBILENET_V1_PRETRAINED_CONFIG_ARCHIVE_MAP, MobileNetV1Config
    from .models.mobilenet_v2 import MOBILENET_V2_PRETRAINED_CONFIG_ARCHIVE_MAP, MobileNetV2Config
    from .models.mobilevit import MOBILEVIT_PRETRAINED_CONFIG_ARCHIVE_MAP, MobileViTConfig
+    from .models.mobilevitv2 import MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP, MobileViTV2Config
    from .models.mpnet import MPNET_PRETRAINED_CONFIG_ARCHIVE_MAP, MPNetConfig, MPNetTokenizer
    from .models.mt5 import MT5Config
    from .models.mvp import MvpConfig, MvpTokenizer
@ -5601,6 +5612,13 @@ if TYPE_CHECKING:
            MobileViTModel,
            MobileViTPreTrainedModel,
        )
+        from .models.mobilevitv2 import (
+            MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST,
+            MobileViTV2ForImageClassification,
+            MobileViTV2ForSemanticSegmentation,
+            MobileViTV2Model,
+            MobileViTV2PreTrainedModel,
+        )
        from .models.mpnet import (
            MPNET_PRETRAINED_MODEL_ARCHIVE_LIST,
            MPNetForMaskedLM,
--- a/src/transformers/models/init.py
+++ b/src/transformers/models/init.py
@ -130,6 +130,7 @@ from . import (
    mobilenet_v1,
    mobilenet_v2,
    mobilevit,
+    mobilevitv2,
    mpnet,
    mt5,
    mvp,
--- a/src/transformers/models/auto/configuration_auto.py
+++ b/src/transformers/models/auto/configuration_auto.py
@ -133,6 +133,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
        ("mobilenet_v1", "MobileNetV1Config"),
        ("mobilenet_v2", "MobileNetV2Config"),
        ("mobilevit", "MobileViTConfig"),
+        ("mobilevitv2", "MobileViTV2Config"),
        ("mpnet", "MPNetConfig"),
        ("mt5", "MT5Config"),
        ("mvp", "MvpConfig"),
@ -319,6 +320,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
        ("mobilenet_v1", "MOBILENET_V1_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("mobilenet_v2", "MOBILENET_V2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("mobilevit", "MOBILEVIT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("mobilevitv2", "MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("mpnet", "MPNET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("mvp", "MVP_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("nat", "NAT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@ -520,6 +522,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
        ("mobilenet_v1", "MobileNetV1"),
        ("mobilenet_v2", "MobileNetV2"),
        ("mobilevit", "MobileViT"),
+        ("mobilevitv2", "MobileViTV2"),
        ("mpnet", "MPNet"),
        ("mt5", "MT5"),
        ("mvp", "MVP"),
--- a/src/transformers/models/auto/image_processing_auto.py
+++ b/src/transformers/models/auto/image_processing_auto.py
@ -77,6 +77,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
        ("mobilenet_v2", "MobileNetV2ImageProcessor"),
        ("mobilevit", "MobileViTImageProcessor"),
        ("mobilevit", "MobileViTImageProcessor"),
+        ("mobilevitv2", "MobileViTImageProcessor"),
        ("nat", "ViTImageProcessor"),
        ("oneformer", "OneFormerImageProcessor"),
        ("owlvit", "OwlViTImageProcessor"),
--- a/src/transformers/models/auto/modeling_auto.py
+++ b/src/transformers/models/auto/modeling_auto.py
@ -131,6 +131,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
        ("mobilenet_v1", "MobileNetV1Model"),
        ("mobilenet_v2", "MobileNetV2Model"),
        ("mobilevit", "MobileViTModel"),
+        ("mobilevitv2", "MobileViTV2Model"),
        ("mpnet", "MPNetModel"),
        ("mt5", "MT5Model"),
        ("mvp", "MvpModel"),
@ -457,6 +458,7 @@ MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
        ("mobilenet_v1", "MobileNetV1ForImageClassification"),
        ("mobilenet_v2", "MobileNetV2ForImageClassification"),
        ("mobilevit", "MobileViTForImageClassification"),
+        ("mobilevitv2", "MobileViTV2ForImageClassification"),
        ("nat", "NatForImageClassification"),
        (
            "perceiver",
@ -496,6 +498,7 @@ MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES = OrderedDict(
        ("dpt", "DPTForSemanticSegmentation"),
        ("mobilenet_v2", "MobileNetV2ForSemanticSegmentation"),
        ("mobilevit", "MobileViTForSemanticSegmentation"),
+        ("mobilevitv2", "MobileViTV2ForSemanticSegmentation"),
        ("segformer", "SegformerForSemanticSegmentation"),
        ("upernet", "UperNetForSemanticSegmentation"),
    ]
--- a/src/transformers/models/mobilevitv2/init.py
+++ b/src/transformers/models/mobilevitv2/init.py
@ -0,0 +1,71 @@
+# 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.
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_torch_available,
+    is_vision_available,
+)
+
+
+_import_structure = {
+    "configuration_mobilevitv2": [
+        "MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "MobileViTV2Config",
+        "MobileViTV2OnnxConfig",
+    ],
+}
+
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_mobilevitv2"] = [
+        "MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "MobileViTV2ForImageClassification",
+        "MobileViTV2ForSemanticSegmentation",
+        "MobileViTV2Model",
+        "MobileViTV2PreTrainedModel",
+    ]
+
+if TYPE_CHECKING:
+    from .configuration_mobilevitv2 import (
+        MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        MobileViTV2Config,
+        MobileViTV2OnnxConfig,
+    )
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_mobilevitv2 import (
+            MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST,
+            MobileViTV2ForImageClassification,
+            MobileViTV2ForSemanticSegmentation,
+            MobileViTV2Model,
+            MobileViTV2PreTrainedModel,
+        )
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
--- a/src/transformers/models/mobilevitv2/configuration_mobilevitv2.py
+++ b/src/transformers/models/mobilevitv2/configuration_mobilevitv2.py
@ -0,0 +1,168 @@
+# coding=utf-8
+# Copyright 2023 The 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.
+""" MobileViTV2 model configuration"""
+
+from collections import OrderedDict
+from typing import Mapping
+
+from packaging import version
+
+from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+MOBILEVITV2_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "apple/mobilevitv2-1.0": "https://huggingface.co/apple/mobilevitv2-1.0/resolve/main/config.json",
+}
+
+
+class MobileViTV2Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`MobileViTV2Model`]. It is used to instantiate a
+    MobileViTV2 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 MobileViTV2
+    [apple/mobilevitv2-1.0](https://huggingface.co/apple/mobilevitv2-1.0) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        num_channels (`int`, *optional*, defaults to 3):
+            The number of input channels.
+        image_size (`int`, *optional*, defaults to 256):
+            The size (resolution) of each image.
+        patch_size (`int`, *optional*, defaults to 2):
+            The size (resolution) of each patch.
+        expand_ratio (`float`, *optional*, defaults to 2.0):
+            Expansion factor for the MobileNetv2 layers.
+        hidden_act (`str` or `function`, *optional*, defaults to `"swish"`):
+            The non-linear activation function (function or string) in the Transformer encoder and convolution layers.
+        conv_kernel_size (`int`, *optional*, defaults to 3):
+            The size of the convolutional kernel in the MobileViTV2 layer.
+        output_stride (`int`, `optional`, defaults to 32):
+            The ratio of the spatial resolution of the output to the resolution of the input image.
+        classifier_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for attached classifiers.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        aspp_out_channels (`int`, `optional`, defaults to 512):
+            Number of output channels used in the ASPP layer for semantic segmentation.
+        atrous_rates (`List[int]`, *optional*, defaults to `[6, 12, 18]`):
+            Dilation (atrous) factors used in the ASPP layer for semantic segmentation.
+        aspp_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the ASPP layer for semantic segmentation.
+        semantic_loss_ignore_index (`int`, *optional*, defaults to 255):
+            The index that is ignored by the loss function of the semantic segmentation model.
+        n_attn_blocks (`List[int]`, *optional*, defaults to `[2, 4, 3]`):
+            The number of attention blocks in each MobileViTV2Layer
+        base_attn_unit_dims (`List[int]`, *optional*, defaults to `[128, 192, 256]`):
+            The base multiplier for dimensions of attention blocks in each MobileViTV2Layer
+        width_multiplier (`float`, *optional*, defaults to 1.0)
+            The width multiplier for MobileViTV2.
+        ffn_multiplier (`int`, *optional*, defaults to 2)
+            The FFN multiplier for MobileViTV2.
+        attn_dropout (`float`, *optional*, defaults to 0.0)
+            The dropout in the attention layer.
+        ffn_dropout (`float`, *optional*, defaults to 0.0)
+            The dropout between FFN layers.
+
+    Example:
+
+    ```python
+    >>> from transformers import MobileViTV2Config, MobileViTV2Model
+
+    >>> # Initializing a mobilevitv2-small style configuration
+    >>> configuration = MobileViTV2Config()
+
+    >>> # Initializing a model from the mobilevitv2-small style configuration
+    >>> model = MobileViTV2Model(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "mobilevitv2"
+
+    def __init__(
+        self,
+        num_channels=3,
+        image_size=256,
+        patch_size=2,
+        expand_ratio=2.0,
+        hidden_act="swish",
+        conv_kernel_size=3,
+        output_stride=32,
+        classifier_dropout_prob=0.1,
+        initializer_range=0.02,
+        layer_norm_eps=1e-5,
+        aspp_out_channels=512,
+        atrous_rates=[6, 12, 18],
+        aspp_dropout_prob=0.1,
+        semantic_loss_ignore_index=255,
+        n_attn_blocks=[2, 4, 3],
+        base_attn_unit_dims=[128, 192, 256],
+        width_multiplier=1.0,
+        ffn_multiplier=2,
+        attn_dropout=0.0,
+        ffn_dropout=0.0,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.expand_ratio = expand_ratio
+        self.hidden_act = hidden_act
+        self.conv_kernel_size = conv_kernel_size
+        self.output_stride = output_stride
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.n_attn_blocks = n_attn_blocks
+        self.base_attn_unit_dims = base_attn_unit_dims
+        self.width_multiplier = width_multiplier
+        self.ffn_multiplier = ffn_multiplier
+        self.ffn_dropout = ffn_dropout
+        self.attn_dropout = attn_dropout
+        self.classifier_dropout_prob = classifier_dropout_prob
+
+        # decode head attributes for semantic segmentation
+        self.aspp_out_channels = aspp_out_channels
+        self.atrous_rates = atrous_rates
+        self.aspp_dropout_prob = aspp_dropout_prob
+        self.semantic_loss_ignore_index = semantic_loss_ignore_index
+
+
+class MobileViTV2OnnxConfig(OnnxConfig):
+    torch_onnx_minimum_version = version.parse("1.11")
+
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        return OrderedDict([("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"})])
+
+    @property
+    def outputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "image-classification":
+            return OrderedDict([("logits", {0: "batch"})])
+        else:
+            return OrderedDict([("last_hidden_state", {0: "batch"}), ("pooler_output", {0: "batch"})])
+
+    @property
+    def atol_for_validation(self) -> float:
+        return 1e-4
--- a/src/transformers/models/mobilevitv2/convert_mlcvnets_to_pytorch.py
+++ b/src/transformers/models/mobilevitv2/convert_mlcvnets_to_pytorch.py
@ -0,0 +1,326 @@
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Convert MobileViTV2 checkpoints from the ml-cvnets library."""
+
+
+import argparse
+import collections
+import json
+from pathlib import Path
+
+import requests
+import torch
+import yaml
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from transformers import (
+    MobileViTImageProcessor,
+    MobileViTV2Config,
+    MobileViTV2ForImageClassification,
+    MobileViTV2ForSemanticSegmentation,
+)
+from transformers.utils import logging
+
+
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
+
+
+def load_orig_config_file(orig_cfg_file):
+    print("Loading config file...")
+
+    def flatten_yaml_as_dict(d, parent_key="", sep="."):
+        items = []
+        for k, v in d.items():
+            new_key = parent_key + sep + k if parent_key else k
+            if isinstance(v, collections.abc.MutableMapping):
+                items.extend(flatten_yaml_as_dict(v, new_key, sep=sep).items())
+            else:
+                items.append((new_key, v))
+        return dict(items)
+
+    config = argparse.Namespace()
+    with open(orig_cfg_file, "r") as yaml_file:
+        try:
+            cfg = yaml.load(yaml_file, Loader=yaml.FullLoader)
+
+            flat_cfg = flatten_yaml_as_dict(cfg)
+            for k, v in flat_cfg.items():
+                setattr(config, k, v)
+        except yaml.YAMLError as exc:
+            logger.error("Error while loading config file: {}. Error message: {}".format(orig_cfg_file, str(exc)))
+    return config
+
+
+def get_mobilevitv2_config(task_name, orig_cfg_file):
+    config = MobileViTV2Config()
+
+    is_segmentation_model = False
+
+    # dataset
+    if task_name.startswith("imagenet1k_"):
+        config.num_labels = 1000
+        if int(task_name.strip().split("_")[-1]) == 384:
+            config.image_size = 384
+        else:
+            config.image_size = 256
+        filename = "imagenet-1k-id2label.json"
+    elif task_name.startswith("imagenet21k_to_1k_"):
+        config.num_labels = 21000
+        if int(task_name.strip().split("_")[-1]) == 384:
+            config.image_size = 384
+        else:
+            config.image_size = 256
+        filename = "imagenet-22k-id2label.json"
+    elif task_name.startswith("ade20k_"):
+        config.num_labels = 151
+        config.image_size = 512
+        filename = "ade20k-id2label.json"
+        is_segmentation_model = True
+    elif task_name.startswith("voc_"):
+        config.num_labels = 21
+        config.image_size = 512
+        filename = "pascal-voc-id2label.json"
+        is_segmentation_model = True
+
+    # orig_config
+    orig_config = load_orig_config_file(orig_cfg_file)
+    assert getattr(orig_config, "model.classification.name", -1) == "mobilevit_v2", "Invalid model"
+    config.width_multiplier = getattr(orig_config, "model.classification.mitv2.width_multiplier", 1.0)
+    assert (
+        getattr(orig_config, "model.classification.mitv2.attn_norm_layer", -1) == "layer_norm_2d"
+    ), "Norm layers other than layer_norm_2d is not supported"
+    config.hidden_act = getattr(orig_config, "model.classification.activation.name", "swish")
+    # config.image_size == getattr(orig_config,  'sampler.bs.crop_size_width', 256)
+
+    if is_segmentation_model:
+        config.output_stride = getattr(orig_config, "model.segmentation.output_stride", 16)
+        if "_deeplabv3" in task_name:
+            config.atrous_rates = getattr(orig_config, "model.segmentation.deeplabv3.aspp_rates", [12, 24, 36])
+            config.aspp_out_channels = getattr(orig_config, "model.segmentation.deeplabv3.aspp_out_channels", 512)
+            config.aspp_dropout_prob = getattr(orig_config, "model.segmentation.deeplabv3.aspp_dropout", 0.1)
+
+    # id2label
+    repo_id = "huggingface/label-files"
+    id2label = json.load(open(hf_hub_download(repo_id, filename, repo_type="dataset"), "r"))
+    id2label = {int(k): v for k, v in id2label.items()}
+    config.id2label = id2label
+    config.label2id = {v: k for k, v in id2label.items()}
+
+    return config
+
+
+def rename_key(dct, old, new):
+    val = dct.pop(old)
+    dct[new] = val
+
+
+def create_rename_keys(state_dict, base_model=False):
+    if base_model:
+        model_prefix = ""
+    else:
+        model_prefix = "mobilevitv2."
+
+    rename_keys = []
+    for k in state_dict.keys():
+        if k[:8] == "encoder.":
+            k_new = k[8:]
+        else:
+            k_new = k
+
+        if ".block." in k:
+            k_new = k_new.replace(".block.", ".")
+        if ".conv." in k:
+            k_new = k_new.replace(".conv.", ".convolution.")
+        if ".norm." in k:
+            k_new = k_new.replace(".norm.", ".normalization.")
+
+        if "conv_1." in k:
+            k_new = k_new.replace("conv_1.", f"{model_prefix}conv_stem.")
+        for i in [1, 2]:
+            if f"layer_{i}." in k:
+                k_new = k_new.replace(f"layer_{i}.", f"{model_prefix}encoder.layer.{i-1}.layer.")
+        if ".exp_1x1." in k:
+            k_new = k_new.replace(".exp_1x1.", ".expand_1x1.")
+        if ".red_1x1." in k:
+            k_new = k_new.replace(".red_1x1.", ".reduce_1x1.")
+
+        for i in [3, 4, 5]:
+            if f"layer_{i}.0." in k:
+                k_new = k_new.replace(f"layer_{i}.0.", f"{model_prefix}encoder.layer.{i-1}.downsampling_layer.")
+            if f"layer_{i}.1.local_rep.0." in k:
+                k_new = k_new.replace(f"layer_{i}.1.local_rep.0.", f"{model_prefix}encoder.layer.{i-1}.conv_kxk.")
+            if f"layer_{i}.1.local_rep.1." in k:
+                k_new = k_new.replace(f"layer_{i}.1.local_rep.1.", f"{model_prefix}encoder.layer.{i-1}.conv_1x1.")
+
+        for i in [3, 4, 5]:
+            if i == 3:
+                j_in = [0, 1]
+            elif i == 4:
+                j_in = [0, 1, 2, 3]
+            elif i == 5:
+                j_in = [0, 1, 2]
+
+            for j in j_in:
+                if f"layer_{i}.1.global_rep.{j}." in k:
+                    k_new = k_new.replace(
+                        f"layer_{i}.1.global_rep.{j}.", f"{model_prefix}encoder.layer.{i-1}.transformer.layer.{j}."
+                    )
+            if f"layer_{i}.1.global_rep.{j+1}." in k:
+                k_new = k_new.replace(
+                    f"layer_{i}.1.global_rep.{j+1}.", f"{model_prefix}encoder.layer.{i-1}.layernorm."
+                )
+
+            if f"layer_{i}.1.conv_proj." in k:
+                k_new = k_new.replace(f"layer_{i}.1.conv_proj.", f"{model_prefix}encoder.layer.{i-1}.conv_projection.")
+
+        if "pre_norm_attn.0." in k:
+            k_new = k_new.replace("pre_norm_attn.0.", "layernorm_before.")
+        if "pre_norm_attn.1." in k:
+            k_new = k_new.replace("pre_norm_attn.1.", "attention.")
+        if "pre_norm_ffn.0." in k:
+            k_new = k_new.replace("pre_norm_ffn.0.", "layernorm_after.")
+        if "pre_norm_ffn.1." in k:
+            k_new = k_new.replace("pre_norm_ffn.1.", "ffn.conv1.")
+        if "pre_norm_ffn.3." in k:
+            k_new = k_new.replace("pre_norm_ffn.3.", "ffn.conv2.")
+
+        if "classifier.1." in k:
+            k_new = k_new.replace("classifier.1.", "classifier.")
+
+        if "seg_head." in k:
+            k_new = k_new.replace("seg_head.", "segmentation_head.")
+        if ".aspp_layer." in k:
+            k_new = k_new.replace(".aspp_layer.", ".")
+        if ".aspp_pool." in k:
+            k_new = k_new.replace(".aspp_pool.", ".")
+
+        rename_keys.append((k, k_new))
+    return rename_keys
+
+
+def remove_unused_keys(state_dict):
+    """remove unused keys (e.g.: seg_head.aux_head)"""
+    keys_to_ignore = []
+    for k in state_dict.keys():
+        if k.startswith("seg_head.aux_head."):
+            keys_to_ignore.append(k)
+    for k in keys_to_ignore:
+        state_dict.pop(k, None)
+
+
+# We will verify our results on an image of cute cats
+def prepare_img():
+    url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+    # url = "https://cdn.britannica.com/86/141086-050-9D7C75EE/Gulfstream-G450-business-jet-passengers.jpg"
+    im = Image.open(requests.get(url, stream=True).raw)
+    return im
+
+
+@torch.no_grad()
+def convert_mobilevitv2_checkpoint(task_name, checkpoint_path, orig_config_path, pytorch_dump_folder_path):
+    """
+    Copy/paste/tweak model's weights to our MobileViTV2 structure.
+    """
+    config = get_mobilevitv2_config(task_name, orig_config_path)
+
+    # load original state_dict
+    checkpoint = torch.load(checkpoint_path, map_location="cpu")
+
+    # load huggingface model
+    if task_name.startswith("ade20k_") or task_name.startswith("voc_"):
+        model = MobileViTV2ForSemanticSegmentation(config).eval()
+        base_model = False
+    else:
+        model = MobileViTV2ForImageClassification(config).eval()
+        base_model = False
+
+    # remove and rename some keys of load the original model
+    state_dict = checkpoint
+    remove_unused_keys(state_dict)
+    rename_keys = create_rename_keys(state_dict, base_model=base_model)
+    for rename_key_src, rename_key_dest in rename_keys:
+        rename_key(state_dict, rename_key_src, rename_key_dest)
+
+    # load modified state_dict
+    model.load_state_dict(state_dict)
+
+    # Check outputs on an image, prepared by MobileViTImageProcessor
+    feature_extractor = MobileViTImageProcessor(crop_size=config.image_size, size=config.image_size + 32)
+    encoding = feature_extractor(images=prepare_img(), return_tensors="pt")
+    outputs = model(**encoding)
+
+    # verify classification model
+    if task_name.startswith("imagenet"):
+        logits = outputs.logits
+        predicted_class_idx = logits.argmax(-1).item()
+        print("Predicted class:", model.config.id2label[predicted_class_idx])
+        if task_name.startswith("imagenet1k_256") and config.width_multiplier == 1.0:
+            # expected_logits for base variant
+            expected_logits = torch.tensor([-1.6336e00, -7.3204e-02, -5.1883e-01])
+            assert torch.allclose(logits[0, :3], expected_logits, atol=1e-4)
+
+    Path(pytorch_dump_folder_path).mkdir(exist_ok=True)
+    print(f"Saving model {task_name} to {pytorch_dump_folder_path}")
+    model.save_pretrained(pytorch_dump_folder_path)
+    print(f"Saving feature extractor to {pytorch_dump_folder_path}")
+    feature_extractor.save_pretrained(pytorch_dump_folder_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--task",
+        default="imagenet1k_256",
+        type=str,
+        help=(
+            "Name of the task for which the MobileViTV2 model you'd like to convert is trained on . "
+            """
+                Classification (ImageNet-1k)
+                    - MobileViTV2 (256x256) : imagenet1k_256
+                    - MobileViTV2 (Trained on 256x256 and Finetuned on 384x384) : imagenet1k_384
+                    - MobileViTV2 (Trained on ImageNet-21k and Finetuned on ImageNet-1k 256x256) :
+                      imagenet21k_to_1k_256
+                    - MobileViTV2 (Trained on ImageNet-21k, Finetuned on ImageNet-1k 256x256, and Finetuned on
+                      ImageNet-1k 384x384) : imagenet21k_to_1k_384
+                Segmentation
+                    - ADE20K Dataset : ade20k_deeplabv3
+                    - Pascal VOC 2012 Dataset: voc_deeplabv3
+            """
+        ),
+        choices=[
+            "imagenet1k_256",
+            "imagenet1k_384",
+            "imagenet21k_to_1k_256",
+            "imagenet21k_to_1k_384",
+            "ade20k_deeplabv3",
+            "voc_deeplabv3",
+        ],
+    )
+
+    parser.add_argument(
+        "--orig_checkpoint_path", required=True, type=str, help="Path to the original state dict (.pt file)."
+    )
+    parser.add_argument("--orig_config_path", required=True, type=str, help="Path to the original config file.")
+    parser.add_argument(
+        "--pytorch_dump_folder_path", required=True, type=str, help="Path to the output PyTorch model directory."
+    )
+
+    args = parser.parse_args()
+    convert_mobilevitv2_checkpoint(
+        args.task, args.orig_checkpoint_path, args.orig_config_path, args.pytorch_dump_folder_path
+    )
--- a/src/transformers/models/mobilevitv2/modeling_mobilevitv2.py
+++ b/src/transformers/models/mobilevitv2/modeling_mobilevitv2.py
--- a/src/transformers/utils/dummy_pt_objects.py
+++ b/src/transformers/utils/dummy_pt_objects.py
@ -4743,6 +4743,37 @@ class MobileViTPreTrainedModel(metaclass=DummyObject):
        requires_backends(self, ["torch"])


+MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class MobileViTV2ForImageClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileViTV2ForSemanticSegmentation(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileViTV2Model(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class MobileViTV2PreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 MPNET_PRETRAINED_MODEL_ARCHIVE_LIST = None


--- a/tests/models/mobilevitv2/init.py
+++ b/tests/models/mobilevitv2/init.py
--- a/tests/models/mobilevitv2/test_modeling_mobilevitv2.py
+++ b/tests/models/mobilevitv2/test_modeling_mobilevitv2.py
@ -0,0 +1,383 @@
+# coding=utf-8
+# Copyright 2023 The 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.
+""" Testing suite for the PyTorch MobileViTV2 model. """
+
+
+import inspect
+import unittest
+
+from transformers import MobileViTV2Config
+from transformers.testing_utils import require_torch, require_vision, slow, torch_device
+from transformers.utils import cached_property, is_torch_available, is_vision_available
+
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import MobileViTV2ForImageClassification, MobileViTV2ForSemanticSegmentation, MobileViTV2Model
+    from transformers.models.mobilevitv2.modeling_mobilevitv2 import (
+        MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST,
+        make_divisible,
+    )
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import MobileViTImageProcessor
+
+
+class MobileViTV2ConfigTester(ConfigTester):
+    def create_and_test_config_common_properties(self):
+        config = self.config_class(**self.inputs_dict)
+        self.parent.assertTrue(hasattr(config, "width_multiplier"))
+
+
+class MobileViTV2ModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        image_size=64,
+        patch_size=2,
+        num_channels=3,
+        hidden_act="swish",
+        conv_kernel_size=3,
+        output_stride=32,
+        classifier_dropout_prob=0.1,
+        initializer_range=0.02,
+        is_training=True,
+        use_labels=True,
+        num_labels=10,
+        scope=None,
+        width_multiplier=0.25,
+        ffn_dropout=0.0,
+        attn_dropout=0.0,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.last_hidden_size = make_divisible(512 * width_multiplier, divisor=8)
+        self.hidden_act = hidden_act
+        self.conv_kernel_size = conv_kernel_size
+        self.output_stride = output_stride
+        self.classifier_dropout_prob = classifier_dropout_prob
+        self.use_labels = use_labels
+        self.is_training = is_training
+        self.num_labels = num_labels
+        self.initializer_range = initializer_range
+        self.scope = scope
+        self.width_multiplier = width_multiplier
+        self.ffn_dropout_prob = ffn_dropout
+        self.attn_dropout_prob = attn_dropout
+
+    def prepare_config_and_inputs(self):
+        pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
+
+        labels = None
+        pixel_labels = None
+        if self.use_labels:
+            labels = ids_tensor([self.batch_size], self.num_labels)
+            pixel_labels = ids_tensor([self.batch_size, self.image_size, self.image_size], self.num_labels)
+
+        config = self.get_config()
+
+        return config, pixel_values, labels, pixel_labels
+
+    def get_config(self):
+        return MobileViTV2Config(
+            image_size=self.image_size,
+            patch_size=self.patch_size,
+            num_channels=self.num_channels,
+            hidden_act=self.hidden_act,
+            conv_kernel_size=self.conv_kernel_size,
+            output_stride=self.output_stride,
+            classifier_dropout_prob=self.classifier_dropout_prob,
+            initializer_range=self.initializer_range,
+            width_multiplier=self.width_multiplier,
+            ffn_dropout=self.ffn_dropout_prob,
+            attn_dropout=self.attn_dropout_prob,
+        )
+
+    def create_and_check_model(self, config, pixel_values, labels, pixel_labels):
+        model = MobileViTV2Model(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values)
+        self.parent.assertEqual(
+            result.last_hidden_state.shape,
+            (
+                self.batch_size,
+                self.last_hidden_size,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+
+    def create_and_check_for_image_classification(self, config, pixel_values, labels, pixel_labels):
+        config.num_labels = self.num_labels
+        model = MobileViTV2ForImageClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values, labels=labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def create_and_check_for_semantic_segmentation(self, config, pixel_values, labels, pixel_labels):
+        config.num_labels = self.num_labels
+        model = MobileViTV2ForSemanticSegmentation(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(pixel_values)
+        self.parent.assertEqual(
+            result.logits.shape,
+            (
+                self.batch_size,
+                self.num_labels,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+        result = model(pixel_values, labels=pixel_labels)
+        self.parent.assertEqual(
+            result.logits.shape,
+            (
+                self.batch_size,
+                self.num_labels,
+                self.image_size // self.output_stride,
+                self.image_size // self.output_stride,
+            ),
+        )
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        config, pixel_values, labels, pixel_labels = config_and_inputs
+        inputs_dict = {"pixel_values": pixel_values}
+        return config, inputs_dict
+
+
+@require_torch
+class MobileViTV2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
+    """
+    Here we also overwrite some of the tests of test_modeling_common.py, as MobileViTV2 does not use input_ids, inputs_embeds,
+    attention_mask and seq_length.
+    """
+
+    all_model_classes = (
+        (MobileViTV2Model, MobileViTV2ForImageClassification, MobileViTV2ForSemanticSegmentation)
+        if is_torch_available()
+        else ()
+    )
+
+    pipeline_model_mapping = (
+        {
+            "feature-extraction": MobileViTV2Model,
+            "image-classification": MobileViTV2ForImageClassification,
+            "image-segmentation": MobileViTV2ForSemanticSegmentation,
+        }
+        if is_torch_available()
+        else {}
+    )
+
+    test_pruning = False
+    test_resize_embeddings = False
+    test_head_masking = False
+    has_attentions = False
+
+    def setUp(self):
+        self.model_tester = MobileViTV2ModelTester(self)
+        self.config_tester = MobileViTV2ConfigTester(self, config_class=MobileViTV2Config, has_text_modality=False)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    @unittest.skip(reason="MobileViTV2 does not use inputs_embeds")
+    def test_inputs_embeds(self):
+        pass
+
+    @unittest.skip(reason="MobileViTV2 does not support input and output embeddings")
+    def test_model_common_attributes(self):
+        pass
+
+    @unittest.skip(reason="MobileViTV2 does not output attentions")
+    def test_attention_outputs(self):
+        pass
+
+    def test_forward_signature(self):
+        config, _ = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+            signature = inspect.signature(model.forward)
+            # signature.parameters is an OrderedDict => so arg_names order is deterministic
+            arg_names = [*signature.parameters.keys()]
+
+            expected_arg_names = ["pixel_values"]
+            self.assertListEqual(arg_names[:1], expected_arg_names)
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_hidden_states_output(self):
+        def check_hidden_states_output(inputs_dict, config, model_class):
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            hidden_states = outputs.hidden_states
+
+            expected_num_stages = 5
+            self.assertEqual(len(hidden_states), expected_num_stages)
+
+            # MobileViTV2's feature maps are of shape (batch_size, num_channels, height, width)
+            # with the width and height being successively divided by 2.
+            divisor = 2
+            for i in range(len(hidden_states)):
+                self.assertListEqual(
+                    list(hidden_states[i].shape[-2:]),
+                    [self.model_tester.image_size // divisor, self.model_tester.image_size // divisor],
+                )
+                divisor *= 2
+
+            self.assertEqual(self.model_tester.output_stride, divisor // 2)
+
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_hidden_states"] = True
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+            # check that output_hidden_states also work using config
+            del inputs_dict["output_hidden_states"]
+            config.output_hidden_states = True
+
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+    def test_for_image_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_image_classification(*config_and_inputs)
+
+    def test_for_semantic_segmentation(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_semantic_segmentation(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in MOBILEVITV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = MobileViTV2Model.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+# We will verify our results on an image of cute cats
+def prepare_img():
+    image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+    return image
+
+
+@require_torch
+@require_vision
+class MobileViTV2ModelIntegrationTest(unittest.TestCase):
+    @cached_property
+    def default_image_processor(self):
+        return (
+            MobileViTImageProcessor.from_pretrained("shehan97/mobilevitv2-1.0-imagenet1k-256")
+            if is_vision_available()
+            else None
+        )
+
+    @slow
+    def test_inference_image_classification_head(self):
+        model = MobileViTV2ForImageClassification.from_pretrained("shehan97/mobilevitv2-1.0-imagenet1k-256").to(
+            torch_device
+        )
+
+        image_processor = self.default_image_processor
+        image = prepare_img()
+        inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
+
+        # forward pass
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        # verify the logits
+        expected_shape = torch.Size((1, 1000))
+        self.assertEqual(outputs.logits.shape, expected_shape)
+
+        expected_slice = torch.tensor([-1.6336e00, -7.3204e-02, -5.1883e-01]).to(torch_device)
+
+        self.assertTrue(torch.allclose(outputs.logits[0, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_inference_semantic_segmentation(self):
+        model = MobileViTV2ForSemanticSegmentation.from_pretrained("shehan97/mobilevitv2-1.0-voc-deeplabv3")
+        model = model.to(torch_device)
+
+        image_processor = MobileViTImageProcessor.from_pretrained("shehan97/mobilevitv2-1.0-voc-deeplabv3")
+
+        image = prepare_img()
+        inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
+
+        # forward pass
+        with torch.no_grad():
+            outputs = model(**inputs)
+        logits = outputs.logits
+
+        # verify the logits
+        expected_shape = torch.Size((1, 21, 32, 32))
+        self.assertEqual(logits.shape, expected_shape)
+
+        expected_slice = torch.tensor(
+            [
+                [[7.0863, 7.1525, 6.8201], [6.6931, 6.8770, 6.8933], [6.2978, 7.0366, 6.9636]],
+                [[-3.7134, -3.6712, -3.6675], [-3.5825, -3.3549, -3.4777], [-3.3435, -3.3979, -3.2857]],
+                [[-2.9329, -2.8003, -2.7369], [-3.0564, -2.4780, -2.0207], [-2.6889, -1.9298, -1.7640]],
+            ],
+            device=torch_device,
+        )
+
+        self.assertTrue(torch.allclose(logits[0, :3, :3, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_post_processing_semantic_segmentation(self):
+        model = MobileViTV2ForSemanticSegmentation.from_pretrained("shehan97/mobilevitv2-1.0-voc-deeplabv3")
+        model = model.to(torch_device)
+
+        image_processor = MobileViTImageProcessor.from_pretrained("shehan97/mobilevitv2-1.0-voc-deeplabv3")
+
+        image = prepare_img()
+        inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
+
+        # forward pass
+        with torch.no_grad():
+            outputs = model(**inputs)
+
+        outputs.logits = outputs.logits.detach().cpu()
+
+        segmentation = image_processor.post_process_semantic_segmentation(outputs=outputs, target_sizes=[(50, 60)])
+        expected_shape = torch.Size((50, 60))
+        self.assertEqual(segmentation[0].shape, expected_shape)
+
+        segmentation = image_processor.post_process_semantic_segmentation(outputs=outputs)
+        expected_shape = torch.Size((32, 32))
+        self.assertEqual(segmentation[0].shape, expected_shape)