Add Nougat (#25942)

* Add conversion script

* Add NougatImageProcessor

* Add crop margin

* More improvements

* Add docs, READMEs

* Remove print statements

* Include model_max_length

* Add NougatTokenizerFast

* Fix imports

* Improve postprocessing

* Improve image processor

* Fix image processor

* Improve normalize method

* More improvements

* More improvements

* Add processor, improve docs

* Simplify fast tokenizer

* Remove test file

* Fix docstrings

* Use NougatProcessor in conversion script

* Add is_levensthein_available

* Add tokenizer tests

* More improvements

* Use numpy instead of opencv

* Add is_cv2_available

* Fix cv2_available

* Add is_nltk_available

* Add image processor tests, improve crop_margin

* Add integration tests

* Improve integration test

* Use do_rescale instead of hacks, thanks Amy

* Remove random_padding

* Address comments

* Address more comments

* Add import

* Address more comments

* Address more comments

* Address comment

* Address comment

* Set max_model_input_sizes

* Add tests

* Add requires_backends

* Add Nougat to exotic tests

* Use to_pil_image

* Address comment regarding nltk

* Add NLTK

* Improve variable names, integration test

* Add test

* refactor, document, and test regexes

* remove named capture groups, add comments

* format

* add non-markdown fixed tokenization

* format

* correct flakyness of args parse

* add regex comments

* test functionalities for crop_image, align long axis and expected output

* add regex tests

* remove cv2 dependency

* test crop_margin equality between cv2 and python

* refactor table regexes to markdown

add newline

* change print to log, improve doc

* fix high count tables correction

* address PR comments: naming, linting, asserts

* Address comments

* Add copied from

* Update conversion script

* Update conversion script to convert both small and base versions

* Add inference example

* Add more info

* Fix style

* Add require annotators to test

* Define all keyword arguments explicitly

* Move cv2 annotator

* Add tokenizer init method

* Transfer checkpoints

* Add reference to Donut

* Address comments

* Skip test

* Remove cv2 method

* Add copied from statements

* Use cached_property

* Fix docstring

* Add file to not doctested

---------
Co-authored-by: Pablo Montalvo <pablo.montalvo.leroux@gmail.com>

.circleci/create_circleci_config.py

@@ -466,13 +466,15 @@ exotic_models_job = CircleCIJob(
         "sudo apt install tesseract-ocr",
         "pip install -U --upgrade-strategy eager pytesseract",
         "pip install -U --upgrade-strategy eager natten",
-        # TODO (ydshieh): Remove this line once `https://github.com/facebookresearch/detectron2/issues/5010` is resolved
-        'pip install -U --upgrade-strategy eager "Pillow<10.0.0"',
+        "pip install -U --upgrade-strategy eager python-Levenshtein",
+        "pip install -U --upgrade-strategy eager opencv-python",
+        "pip install -U --upgrade-strategy eager nltk",
     ],
     tests_to_run=[
         "tests/models/*layoutlmv*",
         "tests/models/*nat",
         "tests/models/deta",
+        "tests/models/nougat",
     ],
     pytest_num_workers=1,
     pytest_options={"durations": 100},

README.md

@@ -426,6 +426,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

README_es.md

@@ -402,6 +402,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

README_hd.md

@@ -374,6 +374,7 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (हुआवेई नूह के आर्क लैब से) साथ में कागज़ [NEZHA: चीनी भाषा समझ के लिए तंत्रिका प्रासंगिक प्रतिनिधित्व](https :/ /arxiv.org/abs/1909.00204) जुन्किउ वेई, ज़ियाओज़े रेन, ज़िआओगुआंग ली, वेनयोंग हुआंग, यी लियाओ, याशेंग वांग, जियाशू लिन, शिन जियांग, जिओ चेन और कुन लियू द्वारा।
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (फ्रॉम मेटा) साथ में पेपर [नो लैंग्वेज लेफ्ट बिहाइंड: स्केलिंग ह्यूमन-सेंटेड मशीन ट्रांसलेशन] (https://arxiv.org/abs/2207.04672) एनएलएलबी टीम द्वारा प्रकाशित।
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta से) the NLLB team. द्वाराअनुसंधान पत्र [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) के साथ जारी किया गया
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (Meta AI से) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. द्वाराअनुसंधान पत्र [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) के साथ जारी किया गया
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (विस्कॉन्सिन विश्वविद्यालय - मैडिसन से) साथ में कागज [Nyströmformer: A Nyström- आधारित एल्गोरिथम आत्म-ध्यान का अनुमान लगाने के लिए ](https://arxiv.org/abs/2102.03902) युनयांग ज़िओंग, झानपेंग ज़ेंग, रुद्रसिस चक्रवर्ती, मिंगक्सिंग टैन, ग्लेन फंग, यिन ली, विकास सिंह द्वारा पोस्ट किया गया।
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs से) पेपर [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) जितेश जैन, जिआचेन ली, मांगटिक चिउ, अली हसनी, निकिता ओरलोव, हम्फ्री शि के द्वारा जारी किया गया है।
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

README_ja.md

@@ -436,6 +436,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab から) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu から公開された研究論文: [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204)
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (Meta から) the NLLB team から公開された研究論文: [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta から) the NLLB team. から公開された研究論文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (Meta AI から) Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. から公開された研究論文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison から) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh から公開された研究論文: [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902)
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs から) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi から公開された研究論文: [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220)
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

README_ko.md

@@ -351,6 +351,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (Huawei Noah’s Ark Lab 에서) Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 의 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 논문과 함께 발표했습니다.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (Meta 에서) the NLLB team 의 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 논문과 함께 발표했습니다.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (Meta 에서 제공)은 the NLLB team.의 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672)논문과 함께 발표했습니다.
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (Meta AI 에서 제공)은 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.의 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418)논문과 함께 발표했습니다.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (the University of Wisconsin - Madison 에서) Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 의 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 논문과 함께 발표했습니다.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (SHI Labs 에서) Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 의 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 논문과 함께 발표했습니다.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

README_zh-hans.md

@@ -375,6 +375,7 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (来自 Meta AI) 伴随论文 [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) 由 Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic 发布。
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (来自 SHI Labs)  伴随论文 [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) 由 Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi 发布。
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (来自 [s-JoL](https://huggingface.co/s-JoL)) 由 [Open-Llama](https://github.com/s-JoL/Open-Llama) 发布.

README_zh-hant.md

@@ -387,6 +387,7 @@ conda install -c huggingface transformers
 1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](https://huggingface.co/docs/transformers/model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](https://huggingface.co/docs/transformers/main/model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](https://huggingface.co/docs/transformers/model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](https://huggingface.co/docs/transformers/model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).

docs/source/en/_toctree.yml

@@ -685,6 +685,8 @@
         title: MatCha
       - local: model_doc/mgp-str
         title: MGP-STR
+      - local: model_doc/nougat
+        title: Nougat
       - local: model_doc/oneformer
         title: OneFormer
       - local: model_doc/owlvit

docs/source/en/index.md

@@ -191,6 +191,7 @@ The documentation is organized into five sections:
 1. **[Nezha](model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
 1. **[NLLB](model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
 1. **[NLLB-MOE](model_doc/nllb-moe)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nougat](model_doc/nougat)** (from Meta AI) released with the paper [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic.
 1. **[Nyströmformer](model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
 1. **[OneFormer](model_doc/oneformer)** (from SHI Labs) released with the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jitesh Jain, Jiachen Li, MangTik Chiu, Ali Hassani, Nikita Orlov, Humphrey Shi.
 1. **[OpenLlama](model_doc/open-llama)** (from [s-JoL](https://huggingface.co/s-JoL)) released in [Open-Llama](https://github.com/s-JoL/Open-Llama).
@@ -411,6 +412,7 @@ Flax), PyTorch, and/or TensorFlow.
 |              NAT              |       ✅        |         ❌         |      ❌      |
 |             Nezha             |       ✅        |         ❌         |      ❌      |
 |           NLLB-MOE            |       ✅        |         ❌         |      ❌      |
+|            Nougat             |       ✅        |         ✅         |      ✅      |
 |         Nyströmformer         |       ✅        |         ❌         |      ❌      |
 |           OneFormer           |       ✅        |         ❌         |      ❌      |
 |          OpenAI GPT           |       ✅        |         ✅         |      ❌      |

docs/source/en/model_doc/nougat.md

+<!--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
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+specific language governing permissions and limitations under the License. -->
+
+# Nougat
+
+## Overview
+
+The Nougat model was proposed in [Nougat: Neural Optical Understanding for Academic Documents](https://arxiv.org/abs/2308.13418) by
+Lukas Blecher, Guillem Cucurull, Thomas Scialom, Robert Stojnic. Nougat uses the same architecture as [Donut](donut), meaning an image Transformer
+encoder and an autoregressive text Transformer decoder to translate scientific PDFs to markdown, enabling easier access to them.
+
+The abstract from the paper is the following:
+
+*Scientific knowledge is predominantly stored in books and scientific journals, often in the form of PDFs. However, the PDF format leads to a loss of semantic information, particularly for mathematical expressions. We propose Nougat (Neural Optical Understanding for Academic Documents), a Visual Transformer model that performs an Optical Character Recognition (OCR) task for processing scientific documents into a markup language, and demonstrate the effectiveness of our model on a new dataset of scientific documents. The proposed approach offers a promising solution to enhance the accessibility of scientific knowledge in the digital age, by bridging the gap between human-readable documents and machine-readable text. We release the models and code to accelerate future work on scientific text recognition.*
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/nougat_architecture.jpg"
+alt="drawing" width="600"/>
+
+<small> Nougat high-level overview. Taken from the <a href="https://arxiv.org/abs/2308.13418">original paper</a>. </small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found
+[here](https://github.com/facebookresearch/nougat).
+
+Tips:
+
+- The quickest way to get started with Nougat is by checking the [tutorial
+  notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Nougat), which show how to use the model
+  at inference time as well as fine-tuning on custom data.
+- Nougat is always used within the [VisionEncoderDecoder](vision-encoder-decoder) framework. The model is identical to [Donut](donut) in terms of architecture.
+
+## Inference
+
+Nougat's [`VisionEncoderDecoder`] model accepts images as input and makes use of
+[`~generation.GenerationMixin.generate`] to autoregressively generate text given the input image.
+
+The [`NougatImageProcessor`] class is responsible for preprocessing the input image and
+[`NougatTokenizerFast`] decodes the generated target tokens to the target string. The
+[`NougatProcessor`] wraps [`NougatImageProcessor`] and [`NougatTokenizerFast`] classes
+into a single instance to both extract the input features and decode the predicted token ids.
+
+- Step-by-step PDF transcription
+
+```py
+>>> from huggingface_hub import hf_hub_download
+>>> import re
+>>> from PIL import Image
+
+>>> from transformers import NougatProcessor, VisionEncoderDecoderModel
+>>> from datasets import load_dataset
+>>> import torch
+
+>>> processor = NougatProcessor.from_pretrained("facebook/nougat-base")
+>>> model = VisionEncoderDecoderModel.from_pretrained("facebook/nougat-base")
+
+>>> device = "cuda" if torch.cuda.is_available() else "cpu"
+>>> model.to(device)  # doctest: +IGNORE_RESULT
+
+>>> # prepare PDF image for the model
+>>> filepath = hf_hub_download(repo_id="hf-internal-testing/fixtures_docvqa", filename="nougat_paper.png", repo_type="dataset")
+>>> image = Image.open(filepath)
+>>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+>>> # generate transcription (here we only generate 30 tokens)
+>>> outputs = model.generate(
+...     pixel_values.to(device),
+...     min_length=1,
+...     max_new_tokens=30,
+...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+... )
+
+>>> sequence = processor.batch_decode(outputs, skip_special_tokens=True)[0]
+>>> sequence = processor.post_process_generation(sequence, fix_markdown=False)
+>>> # note: we're using repr here such for the sake of printing the \n characters, feel free to just print the sequence
+>>> print(repr(sequence))
+'\n\n# Nougat: Neural Optical Understanding for Academic Documents\n\n Lukas Blecher\n\nCorrespondence to: lblecher@'
+```
+
+See the [model hub](https://huggingface.co/models?filter=nougat) to look for Nougat checkpoints.
+
+## NougatImageProcessor
+
+[[autodoc]] NougatImageProcessor
+    - preprocess
+
+## NougatTokenizerFast
+
+[[autodoc]] NougatTokenizerFast
+
+## NougatProcessor
+
+[[autodoc]] NougatProcessor
+    - __call__
+    - from_pretrained
+    - save_pretrained
+    - batch_decode
+    - decode
+    - post_process_generation
\ No newline at end of file

src/transformers/__init__.py

@@ -453,6 +453,7 @@ _import_structure = {
     "models.nezha": ["NEZHA_PRETRAINED_CONFIG_ARCHIVE_MAP", "NezhaConfig"],
     "models.nllb": [],
     "models.nllb_moe": ["NLLB_MOE_PRETRAINED_CONFIG_ARCHIVE_MAP", "NllbMoeConfig"],
+    "models.nougat": ["NougatProcessor"],
     "models.nystromformer": [
         "NYSTROMFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "NystromformerConfig",
@@ -859,6 +860,7 @@ else:
     _import_structure["models.mt5"].append("MT5TokenizerFast")
     _import_structure["models.mvp"].append("MvpTokenizerFast")
     _import_structure["models.nllb"].append("NllbTokenizerFast")
+    _import_structure["models.nougat"].append("NougatTokenizerFast")
     _import_structure["models.openai"].append("OpenAIGPTTokenizerFast")
     _import_structure["models.pegasus"].append("PegasusTokenizerFast")
     _import_structure["models.realm"].append("RealmTokenizerFast")
@@ -973,6 +975,7 @@ else:
     _import_structure["models.mobilenet_v1"].extend(["MobileNetV1FeatureExtractor", "MobileNetV1ImageProcessor"])
     _import_structure["models.mobilenet_v2"].extend(["MobileNetV2FeatureExtractor", "MobileNetV2ImageProcessor"])
     _import_structure["models.mobilevit"].extend(["MobileViTFeatureExtractor", "MobileViTImageProcessor"])
+    _import_structure["models.nougat"].append("NougatImageProcessor")
     _import_structure["models.oneformer"].extend(["OneFormerImageProcessor"])
     _import_structure["models.owlvit"].extend(["OwlViTFeatureExtractor", "OwlViTImageProcessor"])
     _import_structure["models.perceiver"].extend(["PerceiverFeatureExtractor", "PerceiverImageProcessor"])
@@ -4566,6 +4569,7 @@ if TYPE_CHECKING:
     from .models.nat import NAT_PRETRAINED_CONFIG_ARCHIVE_MAP, NatConfig
     from .models.nezha import NEZHA_PRETRAINED_CONFIG_ARCHIVE_MAP, NezhaConfig
     from .models.nllb_moe import NLLB_MOE_PRETRAINED_CONFIG_ARCHIVE_MAP, NllbMoeConfig
+    from .models.nougat import NougatProcessor
     from .models.nystromformer import NYSTROMFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, NystromformerConfig
     from .models.oneformer import ONEFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, OneFormerConfig, OneFormerProcessor
     from .models.openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig, OpenAIGPTTokenizer
@@ -4944,6 +4948,7 @@ if TYPE_CHECKING:
         from .models.mt5 import MT5TokenizerFast
         from .models.mvp import MvpTokenizerFast
         from .models.nllb import NllbTokenizerFast
+        from .models.nougat import NougatTokenizerFast
         from .models.openai import OpenAIGPTTokenizerFast
         from .models.pegasus import PegasusTokenizerFast
         from .models.realm import RealmTokenizerFast
@@ -5029,6 +5034,7 @@ if TYPE_CHECKING:
         from .models.mobilenet_v1 import MobileNetV1FeatureExtractor, MobileNetV1ImageProcessor
         from .models.mobilenet_v2 import MobileNetV2FeatureExtractor, MobileNetV2ImageProcessor
         from .models.mobilevit import MobileViTFeatureExtractor, MobileViTImageProcessor
+        from .models.nougat import NougatImageProcessor
         from .models.oneformer import OneFormerImageProcessor
         from .models.owlvit import OwlViTFeatureExtractor, OwlViTImageProcessor
         from .models.perceiver import PerceiverFeatureExtractor, PerceiverImageProcessor

src/transformers/models/__init__.py

@@ -147,6 +147,7 @@ from . import (
     nezha,
     nllb,
     nllb_moe,
+    nougat,
     nystromformer,
     oneformer,
     openai,

src/transformers/models/auto/configuration_auto.py

@@ -152,6 +152,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("nat", "NatConfig"),
         ("nezha", "NezhaConfig"),
         ("nllb-moe", "NllbMoeConfig"),
+        ("nougat", "VisionEncoderDecoderConfig"),
         ("nystromformer", "NystromformerConfig"),
         ("oneformer", "OneFormerConfig"),
         ("open-llama", "OpenLlamaConfig"),
@@ -580,6 +581,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("nezha", "Nezha"),
         ("nllb", "NLLB"),
         ("nllb-moe", "NLLB-MOE"),
+        ("nougat", "Nougat"),
         ("nystromformer", "Nyströmformer"),
         ("oneformer", "OneFormer"),
         ("open-llama", "OpenLlama"),

src/transformers/models/auto/image_processing_auto.py

@@ -82,6 +82,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("mobilevit", "MobileViTImageProcessor"),
         ("mobilevitv2", "MobileViTImageProcessor"),
         ("nat", "ViTImageProcessor"),
+        ("nougat", "NougatImageProcessor"),
         ("oneformer", "OneFormerImageProcessor"),
         ("owlvit", "OwlViTImageProcessor"),
         ("perceiver", "PerceiverImageProcessor"),

src/transformers/models/donut/image_processing_donut.py

@@ -265,8 +265,7 @@ class DonutImageProcessor(BaseImageProcessor):
         **kwargs,
     ) -> np.ndarray:
         """
-        Resize an image. The shortest edge of the image is resized to size["shortest_edge"], with the longest edge
-        resized to keep the input aspect ratio.
+        Resizes `image` to `(height, width)` specified by `size` using the PIL library.
 
         Args:
             image (`np.ndarray`):

src/transformers/models/nougat/__init__.py

+# 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_tokenizers_available, is_vision_available
+
+
+_import_structure = {
+    "processing_nougat": ["NougatProcessor"],
+}
+
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["tokenization_nougat_fast"] = ["NougatTokenizerFast"]
+
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["image_processing_nougat"] = ["NougatImageProcessor"]
+
+
+if TYPE_CHECKING:
+    from .processing_nougat import NougatProcessor
+
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .tokenization_nougat_fast import NougatTokenizerFast
+
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .image_processing_nougat import NougatImageProcessor
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure)

src/transformers/models/nougat/convert_nougat_to_hf.py

+# 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 Nougat checkpoints using the original `nougat` library. URL:
+https://github.com/facebookresearch/nougat/tree/main"""
+
+import argparse
+
+import torch
+from huggingface_hub import hf_hub_download
+from nougat import NougatModel
+from nougat.dataset.rasterize import rasterize_paper
+from nougat.utils.checkpoint import get_checkpoint
+from PIL import Image
+
+from transformers import (
+    DonutSwinConfig,
+    DonutSwinModel,
+    MBartConfig,
+    MBartForCausalLM,
+    NougatImageProcessor,
+    NougatProcessor,
+    NougatTokenizerFast,
+    VisionEncoderDecoderModel,
+)
+
+
+def get_configs(model):
+    original_config = model.config
+
+    encoder_config = DonutSwinConfig(
+        image_size=original_config.input_size,
+        patch_size=4,
+        depths=original_config.encoder_layer,
+        num_heads=[4, 8, 16, 32],
+        window_size=original_config.window_size,
+        embed_dim=128,
+    )
+    decoder_config = MBartConfig(
+        is_decoder=True,
+        is_encoder_decoder=False,
+        add_cross_attention=True,
+        decoder_layers=original_config.decoder_layer,
+        max_position_embeddings=original_config.max_position_embeddings,
+        vocab_size=len(
+            model.decoder.tokenizer
+        ),  # several special tokens are added to the vocab of XLMRobertaTokenizer, see repo on the hub (added_tokens.json)
+        scale_embedding=True,
+        add_final_layer_norm=True,
+        tie_word_embeddings=False,
+    )
+
+    return encoder_config, decoder_config
+
+
+# Copied from transformers.models.donut.convert_donut_to_pytorch.rename_key
+def rename_key(name):
+    if "encoder.model" in name:
+        name = name.replace("encoder.model", "encoder")
+    if "decoder.model" in name:
+        name = name.replace("decoder.model", "decoder")
+    if "patch_embed.proj" in name:
+        name = name.replace("patch_embed.proj", "embeddings.patch_embeddings.projection")
+    if "patch_embed.norm" in name:
+        name = name.replace("patch_embed.norm", "embeddings.norm")
+    if name.startswith("encoder"):
+        if "layers" in name:
+            name = "encoder." + name
+        if "attn.proj" in name:
+            name = name.replace("attn.proj", "attention.output.dense")
+        if "attn" in name and "mask" not in name:
+            name = name.replace("attn", "attention.self")
+        if "norm1" in name:
+            name = name.replace("norm1", "layernorm_before")
+        if "norm2" in name:
+            name = name.replace("norm2", "layernorm_after")
+        if "mlp.fc1" in name:
+            name = name.replace("mlp.fc1", "intermediate.dense")
+        if "mlp.fc2" in name:
+            name = name.replace("mlp.fc2", "output.dense")
+
+        if name == "encoder.norm.weight":
+            name = "encoder.layernorm.weight"
+        if name == "encoder.norm.bias":
+            name = "encoder.layernorm.bias"
+
+    return name
+
+
+# Copied from transformers.models.donut.convert_donut_to_pytorch.convert_state_dict
+def convert_state_dict(orig_state_dict, model):
+    for key in orig_state_dict.copy().keys():
+        val = orig_state_dict.pop(key)
+
+        if "qkv" in key:
+            key_split = key.split(".")
+            layer_num = int(key_split[3])
+            block_num = int(key_split[5])
+            dim = model.encoder.encoder.layers[layer_num].blocks[block_num].attention.self.all_head_size
+
+            if "weight" in key:
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.query.weight"
+                ] = val[:dim, :]
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.key.weight"
+                ] = val[dim : dim * 2, :]
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.value.weight"
+                ] = val[-dim:, :]
+            else:
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.query.bias"
+                ] = val[:dim]
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.key.bias"
+                ] = val[dim : dim * 2]
+                orig_state_dict[
+                    f"encoder.encoder.layers.{layer_num}.blocks.{block_num}.attention.self.value.bias"
+                ] = val[-dim:]
+        elif "attn_mask" in key or key in ["encoder.model.norm.weight", "encoder.model.norm.bias"]:
+            # HuggingFace implementation doesn't use attn_mask buffer
+            # and model doesn't use final LayerNorms for the encoder
+            pass
+        else:
+            orig_state_dict[rename_key(key)] = val
+
+    return orig_state_dict
+
+
+def convert_nougat_checkpoint(model_tag, pytorch_dump_folder_path=None, push_to_hub=False):
+    # load original model
+    checkpoint_path = get_checkpoint(None, model_tag)
+    original_model = NougatModel.from_pretrained(checkpoint_path)
+    original_model.eval()
+
+    # load HuggingFace model
+    encoder_config, decoder_config = get_configs(original_model)
+    encoder = DonutSwinModel(encoder_config)
+    decoder = MBartForCausalLM(decoder_config)
+    model = VisionEncoderDecoderModel(encoder=encoder, decoder=decoder)
+    model.eval()
+
+    state_dict = original_model.state_dict()
+    new_state_dict = convert_state_dict(state_dict, model)
+    model.load_state_dict(new_state_dict)
+
+    # verify results on PDF
+    filepath = hf_hub_download(repo_id="ysharma/nougat", filename="input/nougat.pdf", repo_type="space")
+    images = rasterize_paper(pdf=filepath, return_pil=True)
+    image = Image.open(images[0])
+
+    tokenizer_file = checkpoint_path / "tokenizer.json"
+    tokenizer = NougatTokenizerFast(tokenizer_file=str(tokenizer_file))
+    tokenizer.pad_token = "<pad>"
+    tokenizer.bos_token = "<s>"
+    tokenizer.eos_token = "</s>"
+    tokenizer.unk_token = "<unk>"
+    tokenizer.model_max_length = original_model.config.max_length
+
+    size = {"height": original_model.config.input_size[0], "width": original_model.config.input_size[1]}
+    image_processor = NougatImageProcessor(
+        do_align_long_axis=original_model.config.align_long_axis,
+        size=size,
+    )
+    processor = NougatProcessor(image_processor=image_processor, tokenizer=tokenizer)
+
+    # verify pixel_values
+    pixel_values = processor(image, return_tensors="pt").pixel_values
+    original_pixel_values = original_model.encoder.prepare_input(image).unsqueeze(0)
+
+    assert torch.allclose(original_pixel_values, pixel_values)
+
+    # verify patch embeddings
+    original_patch_embed = original_model.encoder.model.patch_embed(pixel_values)
+    patch_embeddings, _ = model.encoder.embeddings(pixel_values)
+    assert torch.allclose(original_patch_embed, patch_embeddings)
+
+    # verify encoder hidden states
+    original_last_hidden_state = original_model.encoder(pixel_values)
+    last_hidden_state = model.encoder(pixel_values).last_hidden_state
+    assert torch.allclose(original_last_hidden_state, last_hidden_state, atol=1e-2)
+
+    # NOTE original model does not use tied weights for embeddings of decoder
+    original_embeddings = original_model.decoder.model.model.decoder.embed_tokens
+    embeddings = model.decoder.model.decoder.embed_tokens
+    assert torch.allclose(original_embeddings.weight, embeddings.weight, atol=1e-3)
+
+    # verify decoder hidden states
+    prompt = "hello world"
+    decoder_input_ids = original_model.decoder.tokenizer(
+        prompt, add_special_tokens=False, return_tensors="pt"
+    ).input_ids
+    decoder_attention_mask = torch.ones_like(decoder_input_ids)
+    original_logits = original_model(
+        image_tensors=pixel_values, decoder_input_ids=decoder_input_ids, attention_mask=decoder_attention_mask
+    ).logits
+    logits = model(
+        pixel_values,
+        decoder_input_ids=decoder_input_ids[:, :-1],
+        decoder_attention_mask=decoder_attention_mask[:, :-1],
+    ).logits
+    assert torch.allclose(original_logits, logits, atol=1e-3)
+
+    # verify generation
+    outputs = model.generate(
+        pixel_values,
+        min_length=1,
+        max_length=30,
+        pad_token_id=tokenizer.pad_token_id,
+        eos_token_id=tokenizer.eos_token_id,
+        use_cache=True,
+        bad_words_ids=[
+            [tokenizer.unk_token_id],
+        ],
+        return_dict_in_generate=True,
+        do_sample=False,
+    )
+    generated = tokenizer.batch_decode(outputs.sequences, skip_special_tokens=True)[0]
+
+    if model_tag == "0.1.0-base":
+        expected_generation = "# Nougat: Neural Optical Understanding for Academic Documents\n\nLukas Blecher\n\nCorrespondence to: lblec"
+    elif model_tag == "0.1.0-small":
+        expected_generation = (
+            "# Nougat: Neural Optical Understanding for Academic Documents\n\nLukas Blecher\n\nCorrespondence to: lble"
+        )
+    else:
+        raise ValueError(f"Unexpected model tag: {model_tag}")
+
+    assert generated == expected_generation
+    print("Looks ok!")
+
+    if pytorch_dump_folder_path is not None:
+        print(f"Saving model and processor to {pytorch_dump_folder_path}")
+        model.save_pretrained(pytorch_dump_folder_path)
+        processor.save_pretrained(pytorch_dump_folder_path)
+
+    if push_to_hub:
+        tag_to_name = {"0.1.0-base": "nougat-base", "0.1.0-small": "nougat-small"}
+        model_name = tag_to_name[model_tag]
+
+        model.push_to_hub(f"facebook/{model_name}")
+        processor.push_to_hub(f"facebook/{model_name}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--model_tag",
+        default="0.1.0-base",
+        required=False,
+        type=str,
+        choices=["0.1.0-base", "0.1.0-small"],
+        help="Tag of the original model you'd like to convert.",
+    )
+    parser.add_argument(
+        "--pytorch_dump_folder_path",
+        default=None,
+        required=False,
+        type=str,
+        help="Path to the output PyTorch model directory.",
+    )
+    parser.add_argument(
+        "--push_to_hub",
+        action="store_true",
+        help="Whether or not to push the converted model and processor to the 🤗 hub.",
+    )
+
+    args = parser.parse_args()
+    convert_nougat_checkpoint(args.model_tag, args.pytorch_dump_folder_path, args.push_to_hub)

src/transformers/models/nougat/image_processing_nougat.py

+# 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.
+"""Image processor class for Nougat."""
+
+from typing import Dict, List, Optional, Union
+
+import numpy as np
+
+from ...image_processing_utils import BaseImageProcessor, BatchFeature, get_size_dict
+from ...image_transforms import (
+    get_resize_output_image_size,
+    pad,
+    resize,
+    to_channel_dimension_format,
+    to_pil_image,
+)
+from ...image_utils import (
+    IMAGENET_DEFAULT_MEAN,
+    IMAGENET_DEFAULT_STD,
+    ChannelDimension,
+    ImageInput,
+    PILImageResampling,
+    get_image_size,
+    infer_channel_dimension_format,
+    is_scaled_image,
+    make_list_of_images,
+    to_numpy_array,
+    valid_images,
+)
+from ...utils import TensorType, logging
+from ...utils.import_utils import is_cv2_available, is_vision_available
+
+
+logger = logging.get_logger(__name__)
+
+
+if is_cv2_available():
+    pass
+
+
+if is_vision_available():
+    import PIL
+
+
+class NougatImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a Nougat image processor.
+
+    Args:
+        do_crop_margin (`bool`, *optional*, defaults to `True`):
+            Whether to crop the image margins.
+        do_resize (`bool`, *optional*, defaults to `True`):
+            Whether to resize the image's (height, width) dimensions to the specified `size`. Can be overridden by
+            `do_resize` in the `preprocess` method.
+        size (`Dict[str, int]` *optional*, defaults to `{"height": 896, "width": 672}`):
+            Size of the image after resizing. Can be overridden by `size` in the `preprocess` method.
+        resample (`PILImageResampling`, *optional*, defaults to `PILImageResampling.BILINEAR`):
+            Resampling filter to use if resizing the image. Can be overridden by `resample` in the `preprocess` method.
+        do_thumbnail (`bool`, *optional*, defaults to `True`):
+            Whether to resize the image using thumbnail method.
+        do_align_long_axis (`bool`, *optional*, defaults to `False`):
+            Whether to align the long axis of the image with the long axis of `size` by rotating by 90 degrees.
+        do_pad (`bool`, *optional*, defaults to `True`):
+            Whether to pad the images to the largest image size in the batch.
+        do_rescale (`bool`, *optional*, defaults to `True`):
+            Whether to rescale the image by the specified scale `rescale_factor`. Can be overridden by the `do_rescale`
+            parameter in the `preprocess` method.
+        rescale_factor (`int` or `float`, *optional*, defaults to `1/255`):
+            Scale factor to use if rescaling the image. Can be overridden by the `rescale_factor` parameter in the
+            `preprocess` method.
+        do_normalize (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the image. Can be overridden by `do_normalize` in the `preprocess` method.
+        image_mean (`float` or `List[float]`, *optional*, defaults to `IMAGENET_DEFAULT_MEAN`):
+            Mean to use if normalizing the image. This is a float or list of floats the length of the number of
+            channels in the image. Can be overridden by the `image_mean` parameter in the `preprocess` method.
+        image_std (`float` or `List[float]`, *optional*, defaults to `IMAGENET_DEFAULT_STD`):
+            Image standard deviation.
+    """
+
+    model_input_names = ["pixel_values"]
+
+    def __init__(
+        self,
+        do_crop_margin: bool = True,
+        do_resize: bool = True,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = PILImageResampling.BILINEAR,
+        do_thumbnail: bool = True,
+        do_align_long_axis: bool = False,
+        do_pad: bool = True,
+        do_rescale: bool = True,
+        rescale_factor: Union[int, float] = 1 / 255,
+        do_normalize: bool = True,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        **kwargs,
+    ) -> None:
+        super().__init__(**kwargs)
+
+        size = size if size is not None else {"height": 896, "width": 672}
+        size = get_size_dict(size)
+
+        self.do_crop_margin = do_crop_margin
+        self.do_resize = do_resize
+        self.size = size
+        self.resample = resample
+        self.do_thumbnail = do_thumbnail
+        self.do_align_long_axis = do_align_long_axis
+        self.do_pad = do_pad
+        self.do_rescale = do_rescale
+        self.rescale_factor = rescale_factor
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean if image_mean is not None else IMAGENET_DEFAULT_MEAN
+        self.image_std = image_std if image_std is not None else IMAGENET_DEFAULT_STD
+
+    def python_find_non_zero(self, image: np.array):
+        """This is a reimplementation of a findNonZero function equivalent to cv2."""
+        non_zero_indices = np.column_stack(np.nonzero(image))
+        idxvec = non_zero_indices[:, [1, 0]]
+        idxvec = idxvec.reshape(-1, 1, 2)
+        return idxvec
+
+    def python_bounding_rect(self, coordinates):
+        """This is a reimplementation of a BoundingRect function equivalent to cv2."""
+        min_values = np.min(coordinates, axis=(0, 1)).astype(int)
+        max_values = np.max(coordinates, axis=(0, 1)).astype(int)
+        x_min, y_min = min_values[0], min_values[1]
+        width = max_values[0] - x_min + 1
+        height = max_values[1] - y_min + 1
+        return x_min, y_min, width, height
+
+    def crop_margin(
+        self,
+        image: np.array,
+        gray_threshold: int = 200,
+        data_format: Optional[ChannelDimension] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    ) -> np.array:
+        """
+        Crops the margin of the image. Gray pixels are considered margin (i.e., pixels with a value below the
+        threshold).
+
+        Args:
+            image (`np.array`):
+                The image to be cropped.
+            gray_threshold (`int`, *optional*, defaults to `200`)
+                Value below which pixels are considered to be gray.
+            data_format (`ChannelDimension`, *optional*):
+                The channel dimension format of the output image. If unset, will use the inferred format from the
+                input.
+            input_data_format (`ChannelDimension`, *optional*):
+                The channel dimension format of the input image. If unset, will use the inferred format from the input.
+        """
+        if input_data_format is None:
+            input_data_format = infer_channel_dimension_format(image)
+
+        image = to_pil_image(image, input_data_format=input_data_format)
+        data = np.array(image.convert("L")).astype(np.uint8)
+        max_val = data.max()
+        min_val = data.min()
+        if max_val == min_val:
+            image = np.array(image)
+            image = (
+                to_channel_dimension_format(image, data_format, input_data_format)
+                if data_format is not None
+                else image
+            )
+            return image
+        data = (data - min_val) / (max_val - min_val) * 255
+        gray = data < gray_threshold
+        coords = self.python_find_non_zero(gray)
+        x_min, y_min, width, height = self.python_bounding_rect(coords)
+        image = image.crop((x_min, y_min, x_min + width, y_min + height))
+        image = np.array(image).astype(np.uint8)
+        image = to_channel_dimension_format(image, input_data_format, ChannelDimension.LAST)
+
+        image = (
+            to_channel_dimension_format(image, data_format, input_data_format) if data_format is not None else image
+        )
+
+        return image
+
+    # Copied from transformers.models.donut.image_processing_donut.DonutImageProcessor.align_long_axis
+    def align_long_axis(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    ) -> np.ndarray:
+        """
+        Align the long axis of the image to the longest axis of the specified size.
+
+        Args:
+            image (`np.ndarray`):
+                The image to be aligned.
+            size (`Dict[str, int]`):
+                The size `{"height": h, "width": w}` to align the long axis to.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The data format of the output image. If unset, the same format as the input image is used.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format of the input image. If not provided, it will be inferred.
+
+        Returns:
+            `np.ndarray`: The aligned image.
+        """
+        input_height, input_width = get_image_size(image, channel_dim=input_data_format)
+        output_height, output_width = size["height"], size["width"]
+
+        if (output_width < output_height and input_width > input_height) or (
+            output_width > output_height and input_width < input_height
+        ):
+            image = np.rot90(image, 3)
+
+        if data_format is not None:
+            image = to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format)
+
+        return image
+
+    def pad_image(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    ) -> np.ndarray:
+        """
+        Pad the image to the specified size at the top, bottom, left and right.
+
+        Args:
+            image (`np.ndarray`):
+                The image to be padded.
+            size (`Dict[str, int]`):
+                The size `{"height": h, "width": w}` to pad the image to.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The data format of the output image. If unset, the same format as the input image is used.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format of the input image. If not provided, it will be inferred.
+        """
+        output_height, output_width = size["height"], size["width"]
+        input_height, input_width = get_image_size(image, channel_dim=input_data_format)
+
+        delta_width = output_width - input_width
+        delta_height = output_height - input_height
+
+        pad_top = delta_height // 2
+        pad_left = delta_width // 2
+
+        pad_bottom = delta_height - pad_top
+        pad_right = delta_width - pad_left
+
+        padding = ((pad_top, pad_bottom), (pad_left, pad_right))
+        return pad(image, padding, data_format=data_format, input_data_format=input_data_format)
+
+    # Copied from transformers.models.donut.image_processing_donut.DonutImageProcessor.thumbnail
+    def thumbnail(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        resample: PILImageResampling = PILImageResampling.BICUBIC,
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs,
+    ) -> np.ndarray:
+        """
+        Resize the image to make a thumbnail. The image is resized so that no dimension is larger than any
+        corresponding dimension of the specified size.
+
+        Args:
+            image (`np.ndarray`):
+                The image to be resized.
+            size (`Dict[str, int]`):
+                The size `{"height": h, "width": w}` to resize the image to.
+            resample (`PILImageResampling`, *optional*, defaults to `PILImageResampling.BICUBIC`):
+                The resampling filter to use.
+            data_format (`Optional[Union[str, ChannelDimension]]`, *optional*):
+                The data format of the output image. If unset, the same format as the input image is used.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format of the input image. If not provided, it will be inferred.
+        """
+        input_height, input_width = get_image_size(image, channel_dim=input_data_format)
+        output_height, output_width = size["height"], size["width"]
+
+        # We always resize to the smallest of either the input or output size.
+        height = min(input_height, output_height)
+        width = min(input_width, output_width)
+
+        if height == input_height and width == input_width:
+            return image
+
+        if input_height > input_width:
+            width = int(input_width * height / input_height)
+        elif input_width > input_height:
+            height = int(input_height * width / input_width)
+
+        return resize(
+            image,
+            size=(height, width),
+            resample=resample,
+            reducing_gap=2.0,
+            data_format=data_format,
+            input_data_format=input_data_format,
+            **kwargs,
+        )
+
+    # Copied from transformers.models.donut.image_processing_donut.DonutImageProcessor.resize
+    def resize(
+        self,
+        image: np.ndarray,
+        size: Dict[str, int],
+        resample: PILImageResampling = PILImageResampling.BICUBIC,
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs,
+    ) -> np.ndarray:
+        """
+        Resizes `image` to `(height, width)` specified by `size` using the PIL library.
+
+        Args:
+            image (`np.ndarray`):
+                Image to resize.
+            size (`Dict[str, int]`):
+                Size of the output image.
+            resample (`PILImageResampling`, *optional*, defaults to `PILImageResampling.BICUBIC`):
+                Resampling filter to use when resiizing the image.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format of the image. If not provided, it will be the same as the input image.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format of the input image. If not provided, it will be inferred.
+        """
+        size = get_size_dict(size)
+        shortest_edge = min(size["height"], size["width"])
+        output_size = get_resize_output_image_size(
+            image, size=shortest_edge, default_to_square=False, input_data_format=input_data_format
+        )
+        resized_image = resize(
+            image,
+            size=output_size,
+            resample=resample,
+            data_format=data_format,
+            input_data_format=input_data_format,
+            **kwargs,
+        )
+        return resized_image
+
+    def preprocess(
+        self,
+        images: ImageInput,
+        do_crop_margin: bool = None,
+        do_resize: bool = None,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = None,
+        do_thumbnail: bool = None,
+        do_align_long_axis: bool = None,
+        do_pad: bool = None,
+        do_rescale: bool = None,
+        rescale_factor: Union[int, float] = None,
+        do_normalize: bool = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs,
+    ) -> PIL.Image.Image:
+        """
+        Preprocess an image or batch of images.
+
+        Args:
+            images (`ImageInput`):
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255.
+            do_crop_margin (`bool`, *optional*, defaults to `self.do_crop_margin`):
+                Whether to crop the image margins.
+            do_resize (`bool`, *optional*, defaults to `self.do_resize`):
+                Whether to resize the image.
+            size (`Dict[str, int]`, *optional*, defaults to `self.size`):
+                Size of the image after resizing. Shortest edge of the image is resized to min(size["height"],
+                size["width"]) with the longest edge resized to keep the input aspect ratio.
+            resample (`int`, *optional*, defaults to `self.resample`):
+                Resampling filter to use if resizing the image. This can be one of the enum `PILImageResampling`. Only
+                has an effect if `do_resize` is set to `True`.
+            do_thumbnail (`bool`, *optional*, defaults to `self.do_thumbnail`):
+                Whether to resize the image using thumbnail method.
+            do_align_long_axis (`bool`, *optional*, defaults to `self.do_align_long_axis`):
+                Whether to align the long axis of the image with the long axis of `size` by rotating by 90 degrees.
+            do_pad (`bool`, *optional*, defaults to `self.do_pad`):
+                Whether to pad the images to the largest image size in the batch.
+            do_rescale (`bool`, *optional*, defaults to `self.do_rescale`):
+                Whether to rescale the image by the specified scale `rescale_factor`.
+            rescale_factor (`int` or `float`, *optional*, defaults to `self.rescale_factor`):
+                Scale factor to use if rescaling the image.
+            do_normalize (`bool`, *optional*, defaults to `self.do_normalize`):
+                Whether to normalize the image.
+            image_mean (`float` or `List[float]`, *optional*, defaults to `self.image_mean`):
+                Image mean to use for normalization.
+            image_std (`float` or `List[float]`, *optional*, defaults to `self.image_std`):
+                Image standard deviation to use for normalization.
+            return_tensors (`str` or `TensorType`, *optional*):
+                The type of tensors to return. Can be one of:
+                - Unset: Return a list of `np.ndarray`.
+                - `TensorType.TENSORFLOW` or `'tf'`: Return a batch of type `tf.Tensor`.
+                - `TensorType.PYTORCH` or `'pt'`: Return a batch of type `torch.Tensor`.
+                - `TensorType.NUMPY` or `'np'`: Return a batch of type `np.ndarray`.
+                - `TensorType.JAX` or `'jax'`: Return a batch of type `jax.numpy.ndarray`.
+            data_format (`ChannelDimension` or `str`, *optional*, defaults to `ChannelDimension.FIRST`):
+                The channel dimension format for the output image. Can be one of:
+                - `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - Unset: defaults to the channel dimension format of the input image.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format for the input image. If unset, the channel dimension format is inferred
+                from the input image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - `"none"` or `ChannelDimension.NONE`: image in (height, width) format.
+        """
+        do_crop_margin = do_crop_margin if do_crop_margin is not None else self.do_crop_margin
+        do_resize = do_resize if do_resize is not None else self.do_resize
+        size = size if size is not None else self.size
+        resample = resample if resample is not None else self.resample
+        do_thumbnail = do_thumbnail if do_thumbnail is not None else self.do_thumbnail
+        do_align_long_axis = do_align_long_axis if do_align_long_axis is not None else self.do_align_long_axis
+        do_pad = do_pad if do_pad is not None else self.do_pad
+        do_rescale = do_rescale if do_rescale is not None else self.do_rescale
+        rescale_factor = rescale_factor if rescale_factor is not None else self.rescale_factor
+        do_normalize = do_normalize if do_normalize is not None else self.do_normalize
+        image_mean = image_mean if image_mean is not None else self.image_mean
+        image_std = image_std if image_std is not None else self.image_std
+
+        images = make_list_of_images(images)
+
+        if not valid_images(images):
+            raise ValueError(
+                "Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, "
+                "torch.Tensor, tf.Tensor or jax.ndarray."
+            )
+
+        if do_resize and size is None:
+            raise ValueError("Size must be specified if do_resize is True.")
+
+        if do_pad and size is None:
+            raise ValueError("Size must be specified if do_pad is True.")
+
+        if do_rescale and rescale_factor is None:
+            raise ValueError("Rescale factor must be specified if do_rescale is True.")
+
+        if do_normalize and (image_mean is None or image_std is None):
+            raise ValueError("Image mean and std must be specified if do_normalize is True.")
+
+        # All transformations expect numpy arrays.
+        images = [to_numpy_array(image) for image in images]
+
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+
+        if input_data_format is None:
+            # We assume that all images have the same channel dimension format.
+            input_data_format = infer_channel_dimension_format(images[0])
+
+        if do_crop_margin:
+            images = [self.crop_margin(image, input_data_format=input_data_format) for image in images]
+
+        if do_align_long_axis:
+            images = [self.align_long_axis(image, size=size, input_data_format=input_data_format) for image in images]
+
+        if do_resize:
+            images = [
+                self.resize(image=image, size=size, resample=resample, input_data_format=input_data_format)
+                for image in images
+            ]
+
+        if do_thumbnail:
+            images = [self.thumbnail(image=image, size=size, input_data_format=input_data_format) for image in images]
+
+        if do_pad:
+            images = [self.pad_image(image=image, size=size, input_data_format=input_data_format) for image in images]
+
+        if do_rescale:
+            images = [
+                self.rescale(image=image, scale=rescale_factor, input_data_format=input_data_format)
+                for image in images
+            ]
+
+        if do_normalize:
+            images = [
+                self.normalize(image=image, mean=image_mean, std=image_std, input_data_format=input_data_format)
+                for image in images
+            ]
+
+        images = [
+            to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format) for image in images
+        ]
+
+        data = {"pixel_values": images}
+        return BatchFeature(data=data, tensor_type=return_tensors)

src/transformers/models/nougat/processing_nougat.py

+# 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.
+"""
+Processor class for Nougat.
+"""
+
+from typing import Dict, List, Optional, Union
+
+from transformers.image_utils import ChannelDimension, PILImageResampling
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput, TruncationStrategy
+
+from ...processing_utils import ProcessorMixin
+from ...utils import PaddingStrategy, TensorType
+
+
+class NougatProcessor(ProcessorMixin):
+    r"""
+    Constructs a Nougat processor which wraps a Nougat image processor and a Nougat tokenizer into a single processor.
+
+    [`NougatProcessor`] offers all the functionalities of [`NougatImageProcessor`] and [`NougatTokenizerFast`]. See the
+    [`~NougatProcessor.__call__`] and [`~NougatProcessor.decode`] for more information.
+
+    Args:
+        image_processor ([`NougatImageProcessor`]):
+            An instance of [`NougatImageProcessor`]. The image processor is a required input.
+        tokenizer ([`NougatTokenizerFast`]):
+            An instance of [`NougatTokenizerFast`]. The tokenizer is a required input.
+    """
+    attributes = ["image_processor", "tokenizer"]
+    image_processor_class = "AutoImageProcessor"
+    tokenizer_class = "AutoTokenizer"
+
+    def __init__(self, image_processor, tokenizer):
+        super().__init__(image_processor, tokenizer)
+        self.current_processor = self.image_processor
+
+    def __call__(
+        self,
+        images=None,
+        text=None,
+        do_crop_margin: bool = None,
+        do_resize: bool = None,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = None,
+        do_thumbnail: bool = None,
+        do_align_long_axis: bool = None,
+        do_pad: bool = None,
+        do_rescale: bool = None,
+        rescale_factor: Union[int, float] = None,
+        do_normalize: bool = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        text_pair: Optional[Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]]] = None,
+        text_target: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        text_pair_target: Optional[
+            Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]]
+        ] = None,
+        add_special_tokens: bool = True,
+        padding: Union[bool, str, PaddingStrategy] = False,
+        truncation: Union[bool, str, TruncationStrategy] = None,
+        max_length: Optional[int] = None,
+        stride: int = 0,
+        is_split_into_words: bool = False,
+        pad_to_multiple_of: Optional[int] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        return_token_type_ids: Optional[bool] = None,
+        return_attention_mask: Optional[bool] = None,
+        return_overflowing_tokens: bool = False,
+        return_special_tokens_mask: bool = False,
+        return_offsets_mapping: bool = False,
+        return_length: bool = False,
+        verbose: bool = True,
+    ):
+        if images is None and text is None:
+            raise ValueError("You need to specify either an `images` or `text` input to process.")
+
+        if images is not None:
+            inputs = self.image_processor(
+                images,
+                do_crop_margin=do_crop_margin,
+                do_resize=do_resize,
+                size=size,
+                resample=resample,
+                do_thumbnail=do_thumbnail,
+                do_align_long_axis=do_align_long_axis,
+                do_pad=do_pad,
+                do_rescale=do_rescale,
+                rescale_factor=rescale_factor,
+                do_normalize=do_normalize,
+                image_mean=image_mean,
+                image_std=image_std,
+                return_tensors=return_tensors,
+                data_format=data_format,
+                input_data_format=input_data_format,
+            )
+        if text is not None:
+            encodings = self.tokenizer(
+                text,
+                text_pair=text_pair,
+                text_target=text_target,
+                text_pair_target=text_pair_target,
+                add_special_tokens=add_special_tokens,
+                padding=padding,
+                truncation=truncation,
+                max_length=max_length,
+                stride=stride,
+                is_split_into_words=is_split_into_words,
+                pad_to_multiple_of=pad_to_multiple_of,
+                return_tensors=return_tensors,
+                return_token_type_ids=return_token_type_ids,
+                return_attention_mask=return_attention_mask,
+                return_overflowing_tokens=return_overflowing_tokens,
+                return_special_tokens_mask=return_special_tokens_mask,
+                return_offsets_mapping=return_offsets_mapping,
+                return_length=return_length,
+                verbose=verbose,
+            )
+
+        if text is None:
+            return inputs
+        elif images is None:
+            return encodings
+        else:
+            inputs["labels"] = encodings["input_ids"]
+            return inputs
+
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to NougatTokenizer's [`~PreTrainedTokenizer.batch_decode`]. Please refer
+        to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to NougatTokenizer's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    def post_process_generation(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to NougatTokenizer's [`~PreTrainedTokenizer.post_process_generation`].
+        Please refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.post_process_generation(*args, **kwargs)