Add InstructBLIP (#23460)

* Squash 88 commits

* Use markdown

* Remove mdx files due to bad rebase

* Fix modeling files due to bad rebase

* Fix style

* Update comment

* fix

---------
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

README.md

@@ -375,6 +375,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.

README_es.md

@@ -350,6 +350,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.

README_hd.md

@@ -322,6 +322,7 @@ conda install -c huggingface transformers
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (बर्कले से) साथ में कागज [I-BERT: Integer-only BERT Quantization](https:// arxiv.org/abs/2101.01321) सेहून किम, अमीर घोलमी, ज़ेवेई याओ, माइकल डब्ल्यू महोनी, कर्ट केटज़र द्वारा।
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (Salesforce से) Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi. द्वाराअनुसंधान पत्र [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) के साथ जारी किया गया
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.

README_ja.md

@@ -384,6 +384,7 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (Berkeley から) Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer から公開された研究論文: [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321)
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (OpenAI から) Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever から公開された研究論文: [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/)
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (Salesforce から) Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi. から公開された研究論文 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500)
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (OpenAI から) Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever から公開された研究論文: [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf)
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (Microsoft Research Asia から) Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou から公開された研究論文: [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318)
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (Microsoft Research Asia から) Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou から公開された研究論文: [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740)

README_ko.md

@@ -299,6 +299,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (Berkeley 에서) Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer 의 [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) 논문과 함께 발표했습니다.
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (OpenAI 에서) Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever 의 [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) 논문과 함께 발표했습니다.
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (Salesforce 에서 제공)은 Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.의 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500)논문과 함께 발표했습니다.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (OpenAI 에서) Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever 의 [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) 논문과 함께 발표했습니다.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (Microsoft Research Asia 에서) Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 의 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 논문과 함께 발표했습니다.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (Microsoft Research Asia 에서) Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 의 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 논문과 함께 발표했습니다.

README_zh-hans.md

@@ -323,6 +323,7 @@ conda install -c huggingface transformers
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (来自 Berkeley) 伴随论文 [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) 由 Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer 发布。
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (来自 OpenAI) 伴随论文 [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) 由 Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever 发布。
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (来自 Salesforce) 伴随论文 [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) 由 Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi 发布。
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 由 Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 发布。
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 由 Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 发布。

README_zh-hant.md

@@ -335,6 +335,7 @@ conda install -c huggingface transformers
 1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
 1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
 1. **[Informer](https://huggingface.co/docs/transformers/model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](https://huggingface.co/docs/transformers/main/model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](https://huggingface.co/docs/transformers/model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.

docs/source/en/_toctree.yml

@@ -608,6 +608,8 @@
         title: GIT
       - local: model_doc/groupvit
         title: GroupViT
+      - local: model_doc/instructblip
+        title: InstructBLIP
       - local: model_doc/layoutlm
         title: LayoutLM
       - local: model_doc/layoutlmv2

docs/source/en/index.md

@@ -139,6 +139,7 @@ The documentation is organized into five sections:
 1. **[I-BERT](model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
 1. **[ImageGPT](model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
 1. **[Informer](model_doc/informer)** (from Beihang University, UC Berkeley, Rutgers University, SEDD Company) released with the paper [Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting](https://arxiv.org/abs/2012.07436) by Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang.
+1. **[InstructBLIP](model_doc/instructblip)** (from Salesforce) released with the paper [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
 1. **[Jukebox](model_doc/jukebox)** (from OpenAI) released with the paper [Jukebox: A Generative Model for Music](https://arxiv.org/pdf/2005.00341.pdf) by Prafulla Dhariwal, Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, Ilya Sutskever.
 1. **[LayoutLM](model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[LayoutLMv2](model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
@@ -348,6 +349,7 @@ Flax), PyTorch, and/or TensorFlow.
 |            I-BERT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           ImageGPT            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           Informer            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|         InstructBLIP          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |            Jukebox            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           LayoutLM            |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |          LayoutLMv2           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/instructblip.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
+specific language governing permissions and limitations under the License.
+-->
+
+# InstructBLIP
+
+## Overview
+
+The InstructBLIP model was proposed in [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://arxiv.org/abs/2305.06500) by Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, Steven Hoi.
+InstructBLIP leverages the [BLIP-2](blip2) architecture for visual instruction tuning.
+
+The abstract from the paper is the following:
+
+*General-purpose language models that can solve various language-domain tasks have emerged driven by the pre-training and instruction-tuning pipeline. However, building general-purpose vision-language models is challenging due to the increased task discrepancy introduced by the additional visual input. Although vision-language pre-training has been widely studied, vision-language instruction tuning remains relatively less explored. In this paper, we conduct a systematic and comprehensive study on vision-language instruction tuning based on the pre-trained BLIP-2 models. We gather a wide variety of 26 publicly available datasets, transform them into instruction tuning format and categorize them into two clusters for held-in instruction tuning and held-out zero-shot evaluation. Additionally, we introduce instruction-aware visual feature extraction, a crucial method that enables the model to extract informative features tailored to the given instruction. The resulting InstructBLIP models achieve state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and the larger Flamingo. Our models also lead to state-of-the-art performance when finetuned on individual downstream tasks (e.g., 90.7% accuracy on ScienceQA IMG). Furthermore, we qualitatively demonstrate the advantages of InstructBLIP over concurrent multimodal models.*
+
+Tips:
+
+- InstructBLIP uses the same architecture as [BLIP-2](blip2) with a tiny but important difference: it also feeds the text prompt (instruction) to the Q-Former.
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/instructblip_architecture.jpg"
+alt="drawing" width="600"/>
+
+<small> InstructBLIP architecture. Taken from the <a href="https://arxiv.org/abs/2305.06500">original paper.</a> </small>
+
+This model was contributed by [nielsr](https://huggingface.co/nielsr).
+The original code can be found [here](https://github.com/salesforce/LAVIS/tree/main/projects/instructblip).
+
+
+## InstructBlipConfig
+
+[[autodoc]] InstructBlipConfig
+    - from_vision_qformer_text_configs
+
+## InstructBlipVisionConfig
+
+[[autodoc]] InstructBlipVisionConfig
+
+## InstructBlipQFormerConfig
+
+[[autodoc]] InstructBlipQFormerConfig
+
+## InstructBlipProcessor
+
+[[autodoc]] InstructBlipProcessor
+
+## InstructBlipVisionModel
+
+[[autodoc]] InstructBlipVisionModel
+    - forward
+
+## InstructBlipQFormerModel
+
+[[autodoc]] InstructBlipQFormerModel
+    - forward
+
+## InstructBlipForConditionalGeneration
+
+[[autodoc]] InstructBlipForConditionalGeneration
+    - forward
+    - generate
\ No newline at end of file

src/transformers/__init__.py

@@ -332,6 +332,13 @@ _import_structure = {
     "models.ibert": ["IBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "IBertConfig"],
     "models.imagegpt": ["IMAGEGPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "ImageGPTConfig"],
     "models.informer": ["INFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "InformerConfig"],
+    "models.instructblip": [
+        "INSTRUCTBLIP_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "InstructBlipConfig",
+        "InstructBlipProcessor",
+        "InstructBlipQFormerConfig",
+        "InstructBlipVisionConfig",
+    ],
     "models.jukebox": [
         "JUKEBOX_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "JukeboxConfig",
@@ -1838,6 +1845,15 @@ else:
             "InformerPreTrainedModel",
         ]
     )
+    _import_structure["models.instructblip"].extend(
+        [
+            "INSTRUCTBLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "InstructBlipForConditionalGeneration",
+            "InstructBlipPreTrainedModel",
+            "InstructBlipQFormerModel",
+            "InstructBlipVisionModel",
+        ]
+    )
     _import_structure["models.jukebox"].extend(
         [
             "JUKEBOX_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -4168,6 +4184,13 @@ if TYPE_CHECKING:
     from .models.ibert import IBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, IBertConfig
     from .models.imagegpt import IMAGEGPT_PRETRAINED_CONFIG_ARCHIVE_MAP, ImageGPTConfig
     from .models.informer import INFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, InformerConfig
+    from .models.instructblip import (
+        INSTRUCTBLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        InstructBlipConfig,
+        InstructBlipProcessor,
+        InstructBlipQFormerConfig,
+        InstructBlipVisionConfig,
+    )
     from .models.jukebox import (
         JUKEBOX_PRETRAINED_CONFIG_ARCHIVE_MAP,
         JukeboxConfig,
@@ -5453,6 +5476,13 @@ if TYPE_CHECKING:
             InformerModel,
             InformerPreTrainedModel,
         )
+        from .models.instructblip import (
+            INSTRUCTBLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
+            InstructBlipForConditionalGeneration,
+            InstructBlipPreTrainedModel,
+            InstructBlipQFormerModel,
+            InstructBlipVisionModel,
+        )
         from .models.jukebox import (
             JUKEBOX_PRETRAINED_MODEL_ARCHIVE_LIST,
             JukeboxModel,

src/transformers/models/__init__.py

@@ -100,6 +100,7 @@ from . import (
     ibert,
     imagegpt,
     informer,
+    instructblip,
     jukebox,
     layoutlm,
     layoutlmv2,

src/transformers/models/auto/configuration_auto.py

@@ -107,6 +107,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("ibert", "IBertConfig"),
         ("imagegpt", "ImageGPTConfig"),
         ("informer", "InformerConfig"),
+        ("instructblip", "InstructBlipConfig"),
         ("jukebox", "JukeboxConfig"),
         ("layoutlm", "LayoutLMConfig"),
         ("layoutlmv2", "LayoutLMv2Config"),
@@ -299,6 +300,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("ibert", "IBERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("imagegpt", "IMAGEGPT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("informer", "INFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("instructblip", "INSTRUCTBLIP_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("jukebox", "JUKEBOX_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("layoutlm", "LAYOUTLM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("layoutlmv2", "LAYOUTLMV2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -493,6 +495,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("ibert", "I-BERT"),
         ("imagegpt", "ImageGPT"),
         ("informer", "Informer"),
+        ("instructblip", "InstructBLIP"),
         ("jukebox", "Jukebox"),
         ("layoutlm", "LayoutLM"),
         ("layoutlmv2", "LayoutLMv2"),

src/transformers/models/auto/image_processing_auto.py

@@ -66,6 +66,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("glpn", "GLPNImageProcessor"),
         ("groupvit", "CLIPImageProcessor"),
         ("imagegpt", "ImageGPTImageProcessor"),
+        ("instructblip", "BlipImageProcessor"),
         ("layoutlmv2", "LayoutLMv2ImageProcessor"),
         ("layoutlmv3", "LayoutLMv3ImageProcessor"),
         ("levit", "LevitImageProcessor"),

src/transformers/models/auto/modeling_auto.py

@@ -536,6 +536,7 @@ MODEL_FOR_VISION_2_SEQ_MAPPING_NAMES = OrderedDict(
         ("blip", "BlipForConditionalGeneration"),
         ("blip-2", "Blip2ForConditionalGeneration"),
         ("git", "GitForCausalLM"),
+        ("instructblip", "InstructBlipForConditionalGeneration"),
         ("pix2struct", "Pix2StructForConditionalGeneration"),
         ("vision-encoder-decoder", "VisionEncoderDecoderModel"),
     ]

src/transformers/models/auto/processing_auto.py

@@ -54,6 +54,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("git", "GitProcessor"),
         ("groupvit", "CLIPProcessor"),
         ("hubert", "Wav2Vec2Processor"),
+        ("instructblip", "InstructBlipProcessor"),
         ("layoutlmv2", "LayoutLMv2Processor"),
         ("layoutlmv3", "LayoutLMv3Processor"),
         ("markuplm", "MarkupLMProcessor"),

src/transformers/models/auto/tokenization_auto.py

@@ -167,6 +167,7 @@ else:
             ("herbert", ("HerbertTokenizer", "HerbertTokenizerFast" if is_tokenizers_available() else None)),
             ("hubert", ("Wav2Vec2CTCTokenizer", None)),
             ("ibert", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)),
+            ("instructblip", ("GPT2Tokenizer", "GPT2TokenizerFast" if is_tokenizers_available() else None)),
             ("jukebox", ("JukeboxTokenizer", None)),
             ("layoutlm", ("LayoutLMTokenizer", "LayoutLMTokenizerFast" if is_tokenizers_available() else None)),
             ("layoutlmv2", ("LayoutLMv2Tokenizer", "LayoutLMv2TokenizerFast" if is_tokenizers_available() else None)),

src/transformers/models/blip/modeling_blip.py

@@ -222,9 +222,7 @@ class BlipVisionEmbeddings(nn.Module):
         self.image_size = config.image_size
         self.patch_size = config.patch_size
 
-        self.class_embedding = nn.Parameter(
-            torch.randn(1, 1, self.embed_dim),
-        )
+        self.class_embedding = nn.Parameter(torch.randn(1, 1, self.embed_dim))
 
         self.patch_embedding = nn.Conv2d(
             in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size

src/transformers/models/blip/processing_blip.py

@@ -18,6 +18,7 @@ Processor class for Blip.
 
 from typing import List, Optional, Union
 
+from ...image_utils import ImageInput
 from ...processing_utils import ProcessorMixin
 from ...tokenization_utils_base import BatchEncoding, PaddingStrategy, PreTokenizedInput, TextInput, TruncationStrategy
 from ...utils import TensorType
@@ -47,7 +48,7 @@ class BlipProcessor(ProcessorMixin):
 
     def __call__(
         self,
-        images=None,
+        images: ImageInput = None,
         text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
         add_special_tokens: bool = True,
         padding: Union[bool, str, PaddingStrategy] = False,

src/transformers/models/blip_2/modeling_blip_2.py

@@ -95,9 +95,7 @@ class Blip2VisionEmbeddings(nn.Module):
         self.image_size = config.image_size
         self.patch_size = config.patch_size
 
-        self.class_embedding = nn.Parameter(
-            torch.randn(1, 1, self.embed_dim),
-        )
+        self.class_embedding = nn.Parameter(torch.randn(1, 1, self.embed_dim))
 
         self.patch_embedding = nn.Conv2d(
             in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size
@@ -171,11 +169,7 @@ class Blip2Attention(nn.Module):
         mixed_qkv = mixed_qkv.reshape(bsz, tgt_len, 3, self.num_heads, embed_dim // self.num_heads).permute(
             2, 0, 3, 1, 4
         )
-        query_states, key_states, value_states = (
-            mixed_qkv[0],
-            mixed_qkv[1],
-            mixed_qkv[2],
-        )
+        query_states, key_states, value_states = mixed_qkv[0], mixed_qkv[1], mixed_qkv[2]
 
         # Take the dot product between "query" and "key" to get the raw attention scores.
         attention_scores = torch.matmul(query_states, key_states.transpose(-1, -2))
@@ -1155,11 +1149,7 @@ class Blip2QFormerModel(Blip2PreTrainedModel):
             if type(encoder_hidden_states) == list:
                 encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
             else:
-                (
-                    encoder_batch_size,
-                    encoder_sequence_length,
-                    _,
-                ) = encoder_hidden_states.size()
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
             encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
 
             if type(encoder_attention_mask) == list: