Foundation Transformers

A big convergence of model architectures across language, vision, speech, and multimodal is emerging. However, under the same name "Transformers", the above areas use different implementations for better performance, e.g., Post-LayerNorm for BERT, and Pre-LayerNorm for GPT and vision Transformers. We call for the development of Foundation Transformer for true general-purpose modeling, which serves as a go-to architecture for various tasks and modalities with guaranteed training stability. In this work, we introduce a Transformer variant, named Magneto, to fulfill the goal. Specifically, we propose Sub-LayerNorm for good expressivity, and the initialization strategy theoretically derived from DeepNet for stable scaling up. Extensive experiments demonstrate its superior performance and better stability than the de facto Transformer variants designed for various applications, including language modeling (i.e., BERT, and GPT), machine translation, vision pretraining (i.e., BEiT), speech recognition, and multimodal pretraining (i.e., BEiT-3).

SpeechUT: Bridging Speech and Text with Hidden-Unit for Encoder-Decoder Based Speech-Text Pre-training

The rapid development of single-modal pre-training has prompted researchers to pay more attention to cross-modal pre-training methods. In this paper, we propose a unified-modal speech-unit-text pre-training model, SpeechUT, to connect the representations of a speech encoder and a text decoder with a shared unit encoder. Leveraging hidden-unit as an interface to align speech and text, we can decompose the speech-to-text model into a speech-to-unit model and a unit-to-text model, which can be jointly pre-trained with unpaired speech and text data respectively. Our proposed SpeechUT is fine-tuned and evaluated on automatic speech recognition (ASR) and speech translation (ST) tasks. Experimental results show that SpeechUT gets substantial improvements over strong baselines, and achieves state-of-the-art performance on both the LibriSpeech ASR and MuST-C ST tasks. To better understand the proposed SpeechUT, detailed analyses are conducted. The code and pre-trained models are available at https://aka.ms/SpeechUT.

XDoc: Unified Pre-training for Cross-Format Document Understanding

The surge of pre-training has witnessed the rapid development of document understanding recently. Pre-training and fine-tuning framework has been effectively used to tackle texts in various formats, including plain texts, document texts, and web texts. Despite achieving promising performance, existing pre-trained models usually target one specific document format at one time, making it difficult to combine knowledge from multiple document formats. To address this, we propose XDoc, a unified pre-trained model which deals with different document formats in a single model. For parameter efficiency, we share backbone parameters for different formats such as the word embedding layer and the Transformer layers. Meanwhile, we introduce adaptive layers with lightweight parameters to enhance the distinction across different formats. Experimental results have demonstrated that with only 36.7% parameters, XDoc achieves comparable or even better performance on a variety of downstream tasks compared with the individual pre-trained models, which is cost effective for real-world deployment. The code and pre-trained models will be publicly available at \url{https://aka.ms/xdoc}.

Multilingual Transitivity and Bidirectional Multilingual Agreement for Multilingual Document-level Machine Translation

Multilingual machine translation has been proven an effective strategy to support translation between multiple languages with a single model. However, most studies focus on multilingual sentence translation without considering generating long documents across different languages, which requires an understanding of multilingual context dependency and is typically harder. In this paper, we first spot that naively incorporating auxiliary multilingual data either auxiliary-target or source-auxiliary brings no improvement to the source-target language pair in our interest. Motivated by this observation, we propose a novel framework called Multilingual Transitivity (MTrans) to find an implicit optimal route via source-auxiliary-target within the multilingual model. To encourage MTrans, we propose a novel method called Triplet Parallel Data (TPD), which uses parallel triplets that contain (source-auxiliary, auxiliary-target, and source-target) for training. The auxiliary language then serves as a pivot and automatically facilitates the implicit information transition flow which is easier to translate. We further propose a novel framework called Bidirectional Multilingual Agreement (Bi-MAgree) that encourages the bidirectional agreement between different languages. To encourage Bi-MAgree, we propose a novel method called Multilingual Kullback-Leibler Divergence (MKL) that forces the output distribution of the inputs with the same meaning but in different languages to be consistent with each other. The experimental results indicate that our methods bring consistent improvements over strong baselines on three document translation tasks: IWSLT2015 Zh-En, De-En, and Vi-En. Our analysis validates the usefulness and existence of MTrans and Bi-MAgree, and our frameworks and methods are effective on synthetic auxiliary data.

SpeechLM: Enhanced Speech Pre-Training with Unpaired Textual Data

How to boost speech pre-training with textual data is an unsolved problem due to the fact that speech and text are very different modalities with distinct characteristics. In this paper, we propose a cross-modal Speech and Language Model (SpeechLM) to explicitly align speech and text pre-training with a pre-defined unified discrete representation. Specifically, we introduce two alternative discrete tokenizers to bridge the speech and text modalities, including phoneme-unit and hidden-unit tokenizers, which can be trained using a small amount of paired speech-text data. Based on the trained tokenizers, we convert the unlabeled speech and text data into tokens of phoneme units or hidden units. The pre-training objective is designed to unify the speech and the text into the same discrete semantic space with a unified Transformer network. Leveraging only 10K text sentences, our SpeechLM gets a 16\% relative WER reduction over the best base model performance (from 6.8 to 5.7) on the public LibriSpeech ASR benchmark. Moreover, SpeechLM with fewer parameters even outperforms previous SOTA models on CoVoST-2 speech translation tasks. We also evaluate our SpeechLM on various spoken language processing tasks under the universal representation evaluation framework SUPERB, demonstrating significant improvements on content-related tasks. Our code and models are available at https://aka.ms/SpeechLM.

Towards Multilingual Transitivity and Bidirectional Multilingual Agreement for Multilingual Document-level Machine Translation

Multilingual machine translation has been proven an effective strategy to support translation between multiple languages with a single model. However, most studies focus on multilingual sentence translation without considering generating long documents across different languages, which requires an understanding of multilingual context dependency and is typically harder. In this paper, we first spot that naively incorporating auxiliary multilingual data either auxiliary-target or source-auxiliary brings no improvement to the source-target language pair in our interest. Motivated by this observation, we propose a novel framework called Multilingual Transitivity (MTrans) to find an implicit optimal route via source-auxiliary-target within the multilingual model. To encourage MTrans, we propose a novel method called Triplet Parallel Data (TPD), which uses parallel triplets that contain (source-auxiliary, auxiliary-target, and source-target) for training. The auxiliary language then serves as a pivot and automatically facilitates the implicit information transition flow which is easier to translate. We further propose a novel framework called Bidirectional Multilingual Agreement (Bi-MAgree) that encourages the bidirectional agreement between different languages. To encourage Bi-MAgree, we propose a novel method called Multilingual Kullback-Leibler Divergence (MKL) that forces the output distribution of the inputs with the same meaning but in different languages to be consistent with each other. The experimental results indicate that our methods bring consistent improvements over strong baselines on three document translation tasks: IWSLT2015 Zh-En, De-En, and Vi-En. Our analysis validates the usefulness and existence of MTrans and Bi-MAgree, and our frameworks and methods are effective on synthetic auxiliary data.

Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Tasks

A big convergence of language, vision, and multimodal pretraining is emerging. In this work, we introduce a general-purpose multimodal foundation model BEiT-3, which achieves state-of-the-art transfer performance on both vision and vision-language tasks. Specifically, we advance the big convergence from three aspects: backbone architecture, pretraining task, and model scaling up. We introduce Multiway Transformers for general-purpose modeling, where the modular architecture enables both deep fusion and modality-specific encoding. Based on the shared backbone, we perform masked "language" modeling on images (Imglish), texts (English), and image-text pairs ("parallel sentences") in a unified manner. Experimental results show that BEiT-3 obtains state-of-the-art performance on object detection (COCO), semantic segmentation (ADE20K), image classification (ImageNet), visual reasoning (NLVR2), visual question answering (VQAv2), image captioning (COCO), and cross-modal retrieval (Flickr30K, COCO).

BEiT v2: Masked Image Modeling with Vector-Quantized Visual Tokenizers

Masked image modeling (MIM) has demonstrated impressive results in self-supervised representation learning by recovering corrupted image patches. However, most methods still operate on low-level image pixels, which hinders the exploitation of high-level semantics for representation models. In this study, we propose to use a semantic-rich visual tokenizer as the reconstruction target for masked prediction, providing a systematic way to promote MIM from pixel-level to semantic-level. Specifically, we introduce vector-quantized knowledge distillation to train the tokenizer, which discretizes a continuous semantic space to compact codes. We then pretrain vision Transformers by predicting the original visual tokens for the masked image patches. Moreover, we encourage the model to explicitly aggregate patch information into a global image representation, which facilities linear probing. Experiments on image classification and semantic segmentation show that our approach outperforms all compared MIM methods. On ImageNet-1K (224 size), the base-size BEiT v2 achieves 85.5% top-1 accuracy for fine-tuning and 80.1% top-1 accuracy for linear probing. The large-size BEiT v2 obtains 87.3% top-1 accuracy for ImageNet-1K (224 size) fine-tuning, and 56.7% mIoU on ADE20K for semantic segmentation. The code and pretrained models are available at https://aka.ms/beit.

Learning Diverse Document Representations with Deep Query Interactions for Dense Retrieval

In this paper, we propose a new dense retrieval model which learns diverse document representations with deep query interactions. Our model encodes each document with a set of generated pseudo-queries to get query-informed, multi-view document representations. It not only enjoys high inference efficiency like the vanilla dual-encoder models, but also enables deep query-document interactions in document encoding and provides multi-faceted representations to better match different queries. Experiments on several benchmarks demonstrate the effectiveness of the proposed method, out-performing strong dual encoder baselines.The code is available at \url{https://github.com/jordane95/dual-cross-encoder