Key-value (KV) caching is an important technique to accelerate the inference of large language models (LLMs), but incurs significant memory overhead. To compress the size of KV cache, existing methods often compromise precision or require extra data for calibration, limiting their practicality in LLM deployment. In this paper, we introduce DecoQuant, a novel data-free low-bit quantization technique based on tensor decomposition methods, to effectively compress KV cache. Our core idea is to adjust the outlier distribution of the original matrix by performing tensor decomposition, so that the quantization difficulties are migrated from the matrix to decomposed local tensors. Specially, we find that outliers mainly concentrate on small local tensors, while large tensors tend to have a narrower value range. Based on this finding, we propose to apply low-bit quantization to the large tensor, while maintaining high-precision representation for the small tensor. Furthermore, we utilize the proposed quantization method to compress the KV cache of LLMs to accelerate the inference, and develop an efficient dequantization kernel tailored specifically for DecoQuant. Through extensive experiments, DecoQuant demonstrates remarkable efficiency gains, showcasing up to a 75% reduction in memory footprint while maintaining comparable generation quality.
Multi-modal large language models (MLLMs) have demonstrated remarkable performance across various tasks. However, these models often prioritize capturing global information and overlook the importance of perceiving local information. This limitation hinders their ability to effectively understand fine-grained details and handle grounding tasks that necessitate nuanced comprehension. Although some recent works have made strides in this, they have primarily focused on single-modality inputs. Therefore, we propose GroundingGPT, an end-to-end language enhanced multi-modal grounding model. It is designed to perform fine-grained grounding tasks for three modalities: image, video and audio. To enhance the model’s performance, we adopt a coarse-to-fine training strategy, utilizing a three-stage training approach to progressively enhance the model’s semantic awareness and fine-grained understanding capabilities. Additionally, we employ a diversified stage-specific dataset construction pipeline, developing a multi-modal, multi-granularity dataset tailored for training the model in different stages. Extensive experiments conducted on multiple multi-modal benchmarks demonstrate that our model achieves impressive fine-grained understanding of multi-modal inputs on grounding tasks while maintaining or improving its global comprehension capabilities. Our code, model, and dataset are available at https://github.com/lzw-lzw/GroundingGPT.
Automatic metrics play a crucial role in machine translation. Despite the widespread use of n-gram-based metrics, there has been a recent surge in the development of pre-trained model-based metrics that focus on measuring sentence semantics. However, these neural metrics, while achieving higher correlations with human evaluations, are often considered to be black boxes with potential biases that are difficult to detect. In this study, we systematically analyze and compare various mainstream and cutting-edge automatic metrics from the perspective of their guidance for training machine translation systems. Through Minimum Risk Training (MRT), we find that certain metrics exhibit robustness defects, such as the presence of universal adversarial translations in BLEURT and BARTScore. In-depth analysis suggests two main causes of these robustness deficits: distribution biases in the training datasets, and the tendency of the metric paradigm. By incorporating token-level constraints, we enhance the robustness of evaluation metrics, which in turn leads to an improvement in the performance of machine translation systems. Codes are available at https://github.com/powerpuffpomelo/fairseq_mrt.
In speech translation, leveraging multimodal data to improve model performance and address limitations of individual modalities has shown significant effectiveness. In this paper, we harness the complementary strengths of speech and text to improve speech translation. However, speech and text are disparate modalities, we observe three aspects of modality gap that impede their integration in a speech translation model. To tackle these gaps, we propose **Fuse**-**S**peech-**T**ext (**FuseST**), a cross-modal model which supports three distinct input modalities for translation: speech, text and fused speech-text. We leverage multiple techniques for cross-modal alignment and conduct a comprehensive analysis to assess its impact on speech translation, machine translation and fused speech-text translation. We evaluate FuseST on MuST-C, GigaST and newstest benchmark. Experiments show that the proposed FuseST achieves an average 34.0 BLEU on MuST-C En→De/Es/Fr (vs SOTA +1.1 BLEU). Further experiments demonstrate that FuseST does not degrade on MT task, as observed in previous works. Instead, it yields an average improvement of 3.2 BLEU over the pre-trained MT model. Code is available at https://github.com/WenbiaoYin/FuseST.
With the rise of the metaverse, immersive multimodal conversation has attracted more and more researchers’ attention. Multimodal contexts will become more important for human-computer interaction in the metaverse, especially in shopping domain. Unlike traditional conversation tasks, immersive multimodal conversation has challenges such as multimodal ambiguous candidate identification and multimodal coreference resolution, which makes it more difficult to dialog state tracking and response generation, as described in SIMMC 2.1 challenge, a part of DSTC11. In particular, as the number of objects in the scene increases, the difficulty will increase dramatically. We proposed a prompt-based multi-task learning Encoder-Decoder, in which different subtasks use different prompts to make the model tend to focus on the current subtask. We achieve the winner in ambiguous candidates indentification and runner-up in multimodal coreference resolution (MM-Coref), multimodal dialog state tracking (MM-DST) and assistant response generation. Our code and model are made publicly available at https://github.com/scutcyr/dstc11-simmc2.1-scut-bds-lab.
Recently, Multi-modal Named Entity Recognition (MNER) has attracted a lot of attention. Most of the work utilizes image information through region-level visual representations obtained from a pretrained object detector and relies on an attention mechanism to model the interactions between image and text representations. However, it is difficult to model such interactions as image and text representations are trained separately on the data of their respective modality and are not aligned in the same space. As text representations take the most important role in MNER, in this paper, we propose Image-text Alignments (ITA) to align image features into the textual space, so that the attention mechanism in transformer-based pretrained textual embeddings can be better utilized. ITA first aligns the image into regional object tags, image-level captions and optical characters as visual contexts, concatenates them with the input texts as a new cross-modal input, and then feeds it into a pretrained textual embedding model. This makes it easier for the attention module of a pretrained textual embedding model to model the interaction between the two modalities since they are both represented in the textual space. ITA further aligns the output distributions predicted from the cross-modal input and textual input views so that the MNER model can be more practical in dealing with text-only inputs and robust to noises from images. In our experiments, we show that ITA models can achieve state-of-the-art accuracy on multi-modal Named Entity Recognition datasets, even without image information.
Medical code prediction from clinical notes aims at automatically associating medical codes with the clinical notes. Rare code problem, the medical codes with low occurrences, is prominent in medical code prediction. Recent studies employ deep neural networks and the external knowledge to tackle it. However, such approaches lack interpretability which is a vital issue in medical application. Moreover, due to the lengthy and noisy clinical notes, such approaches fail to achieve satisfactory results. Therefore, in this paper, we propose a novel framework based on medical concept driven attention to incorporate external knowledge for explainable medical code prediction. In specific, both the clinical notes and Wikipedia documents are aligned into topic space to extract medical concepts using topic modeling. Then, the medical concept-driven attention mechanism is applied to uncover the medical code related concepts which provide explanations for medical code prediction. Experimental results on the benchmark dataset show the superiority of the proposed framework over several state-of-the-art baselines.
The MultiCoNER shared task aims at detecting semantically ambiguous and complex named entities in short and low-context settings for multiple languages. The lack of contexts makes the recognition of ambiguous named entities challenging. To alleviate this issue, our team DAMO-NLP proposes a knowledge-based system, where we build a multilingual knowledge base based on Wikipedia to provide related context information to the named entity recognition (NER) model. Given an input sentence, our system effectively retrieves related contexts from the knowledge base. The original input sentences are then augmented with such context information, allowing significantly better contextualized token representations to be captured. Our system wins 10 out of 13 tracks in the MultiCoNER shared task.
Knowledge distillation is a critical technique to transfer knowledge between models, typically from a large model (the teacher) to a more fine-grained one (the student). The objective function of knowledge distillation is typically the cross-entropy between the teacher and the student’s output distributions. However, for structured prediction problems, the output space is exponential in size; therefore, the cross-entropy objective becomes intractable to compute and optimize directly. In this paper, we derive a factorized form of the knowledge distillation objective for structured prediction, which is tractable for many typical choices of the teacher and student models. In particular, we show the tractability and empirical effectiveness of structural knowledge distillation between sequence labeling and dependency parsing models under four different scenarios: 1) the teacher and student share the same factorization form of the output structure scoring function; 2) the student factorization produces more fine-grained substructures than the teacher factorization; 3) the teacher factorization produces more fine-grained substructures than the student factorization; 4) the factorization forms from the teacher and the student are incompatible.
Recent advances in Named Entity Recognition (NER) show that document-level contexts can significantly improve model performance. In many application scenarios, however, such contexts are not available. In this paper, we propose to find external contexts of a sentence by retrieving and selecting a set of semantically relevant texts through a search engine, with the original sentence as the query. We find empirically that the contextual representations computed on the retrieval-based input view, constructed through the concatenation of a sentence and its external contexts, can achieve significantly improved performance compared to the original input view based only on the sentence. Furthermore, we can improve the model performance of both input views by Cooperative Learning, a training method that encourages the two input views to produce similar contextual representations or output label distributions. Experiments show that our approach can achieve new state-of-the-art performance on 8 NER data sets across 5 domains.
Pretrained contextualized embeddings are powerful word representations for structured prediction tasks. Recent work found that better word representations can be obtained by concatenating different types of embeddings. However, the selection of embeddings to form the best concatenated representation usually varies depending on the task and the collection of candidate embeddings, and the ever-increasing number of embedding types makes it a more difficult problem. In this paper, we propose Automated Concatenation of Embeddings (ACE) to automate the process of finding better concatenations of embeddings for structured prediction tasks, based on a formulation inspired by recent progress on neural architecture search. Specifically, a controller alternately samples a concatenation of embeddings, according to its current belief of the effectiveness of individual embedding types in consideration for a task, and updates the belief based on a reward. We follow strategies in reinforcement learning to optimize the parameters of the controller and compute the reward based on the accuracy of a task model, which is fed with the sampled concatenation as input and trained on a task dataset. Empirical results on 6 tasks and 21 datasets show that our approach outperforms strong baselines and achieves state-of-the-art performance with fine-tuned embeddings in all the evaluations.
In structured prediction problems, cross-lingual transfer learning is an efficient way to train quality models for low-resource languages, and further improvement can be obtained by learning from multiple source languages. However, not all source models are created equal and some may hurt performance on the target language. Previous work has explored the similarity between source and target sentences as an approximate measure of strength for different source models. In this paper, we propose a multi-view framework, by leveraging a small number of labeled target sentences, to effectively combine multiple source models into an aggregated source view at different granularity levels (language, sentence, or sub-structure), and transfer it to a target view based on a task-specific model. By encouraging the two views to interact with each other, our framework can dynamically adjust the confidence level of each source model and improve the performance of both views during training. Experiments for three structured prediction tasks on sixteen data sets show that our framework achieves significant improvement over all existing approaches, including these with access to additional source language data.
Zero-shot sequence labeling aims to build a sequence labeler without human-annotated datasets. One straightforward approach is utilizing existing systems (source models) to generate pseudo-labeled datasets and train a target sequence labeler accordingly. However, due to the gap between the source and the target languages/domains, this approach may fail to recover the true labels. In this paper, we propose a novel unified framework for zero-shot sequence labeling with minimum risk training and design a new decomposable risk function that models the relations between the predicted labels from the source models and the true labels. By making the risk function trainable, we draw a connection between minimum risk training and latent variable model learning. We propose a unified learning algorithm based on the expectation maximization (EM) algorithm. We extensively evaluate our proposed approaches on cross-lingual/domain sequence labeling tasks over twenty-one datasets. The results show that our approaches outperform state-of-the-art baseline systems.
Automatic translation of dialogue texts is a much needed demand in many real life scenarios. However, the currently existing neural machine translation delivers unsatisfying results. In this paper, we conduct a deep analysis of a dialogue corpus and summarize three major issues on dialogue translation, including pronoun dropping (), punctuation dropping (), and typos (). In response to these challenges, we propose a joint learning method to identify omission and typo, and utilize context to translate dialogue utterances. To properly evaluate the performance, we propose a manually annotated dataset with 1,931 Chinese-English parallel utterances from 300 dialogues as a benchmark testbed for dialogue translation. Our experiments show that the proposed method improves translation quality by 3.2 BLEU over the baselines. It also elevates the recovery rate of omitted pronouns from 26.09% to 47.16%. We will publish the code and dataset publicly at https://xxx.xx.
This paper describes the systems submitted to IWSLT 2021 by the Volctrans team. We participate in the offline speech translation and text-to-text simultaneous translation tracks. For offline speech translation, our best end-to-end model achieves 7.9 BLEU improvements over the benchmark on the MuST-C test set and is even approaching the results of a strong cascade solution. For text-to-text simultaneous translation, we explore the best practice to optimize the wait-k model. As a result, our final submitted systems exceed the benchmark at around 7 BLEU on the same latency regime. We release our code and model to facilitate both future research works and industrial applications.
This paper presents Self-correcting Encoding (Secoco), a framework that effectively deals with noisy input for robust neural machine translation by introducing self-correcting predictors. Different from previous robust approaches, Secoco enables NMT to explicitly correct noisy inputs and delete specific errors simultaneously with the translation decoding process. Secoco is able to achieve significant improvements over strong baselines on two real-world test sets and a benchmark WMT dataset with good interpretability. We will make our code and dataset publicly available soon.
Entity retrieval, which aims at disambiguating mentions to canonical entities from massive KBs, is essential for many tasks in natural language processing. Recent progress in entity retrieval shows that the dual-encoder structure is a powerful and efficient framework to nominate candidates if entities are only identified by descriptions. However, they ignore the property that meanings of entity mentions diverge in different contexts and are related to various portions of descriptions, which are treated equally in previous works. In this work, we propose Multi-View Entity Representations (MuVER), a novel approach for entity retrieval that constructs multi-view representations for entity descriptions and approximates the optimal view for mentions via a heuristic searching method. Our method achieves the state-of-the-art performance on ZESHEL and improves the quality of candidates on three standard Entity Linking datasets.
Adapting word order from one language to another is a key problem in cross-lingual structured prediction. Current sentence encoders (e.g., RNN, Transformer with position embeddings) are usually word order sensitive. Even with uniform word form representations (MUSE, mBERT), word order discrepancies may hurt the adaptation of models. In this paper, we build structured prediction models with bag-of-words inputs, and introduce a new reordering module to organizing words following the source language order, which learns task-specific reordering strategies from a general-purpose order predictor model. Experiments on zero-shot cross-lingual dependency parsing, POS tagging, and morphological tagging show that our model can significantly improve target language performances, especially for languages that are distant from the source language.
Transition systems usually contain various dynamic structures (e.g., stacks, buffers). An ideal transition-based model should encode these structures completely and efficiently. Previous works relying on templates or neural network structures either only encode partial structure information or suffer from computation efficiency. In this paper, we propose a novel attention-based encoder unifying representation of all structures in a transition system. Specifically, we separate two views of items on structures, namely structure-invariant view and structure-dependent view. With the help of parallel-friendly attention network, we are able to encoding transition states with O(1) additional complexity (with respect to basic feature extractors). Experiments on the PTB and UD show that our proposed method significantly improves the test speed and achieves the best transition-based model, and is comparable to state-of-the-art methods.
Multilingual sequence labeling is a task of predicting label sequences using a single unified model for multiple languages. Compared with relying on multiple monolingual models, using a multilingual model has the benefit of a smaller model size, easier in online serving, and generalizability to low-resource languages. However, current multilingual models still underperform individual monolingual models significantly due to model capacity limitations. In this paper, we propose to reduce the gap between monolingual models and the unified multilingual model by distilling the structural knowledge of several monolingual models (teachers) to the unified multilingual model (student). We propose two novel KD methods based on structure-level information: (1) approximately minimizes the distance between the student’s and the teachers’ structure-level probability distributions, (2) aggregates the structure-level knowledge to local distributions and minimizes the distance between two local probability distributions. Our experiments on 4 multilingual tasks with 25 datasets show that our approaches outperform several strong baselines and have stronger zero-shot generalizability than both the baseline model and teacher models.
The neural linear-chain CRF model is one of the most widely-used approach to sequence labeling. In this paper, we investigate a series of increasingly expressive potential functions for neural CRF models, which not only integrate the emission and transition functions, but also explicitly take the representations of the contextual words as input. Our extensive experiments show that the decomposed quadrilinear potential function based on the vector representations of two neighboring labels and two neighboring words consistently achieves the best performance.
Does neural machine translation yield translations that are congenial with common sense? In this paper, we present a test suite to evaluate the commonsense reasoning capability of neural machine translation. The test suite consists of three test sets, covering lexical and contextless/contextual syntactic ambiguity that requires commonsense knowledge to resolve. We manually create 1,200 triples, each of which contain a source sentence and two contrastive translations, involving 7 different common sense types. Language models pretrained on large-scale corpora, such as BERT, GPT-2, achieve a commonsense reasoning accuracy of lower than 72% on target translations of this test suite. We conduct extensive experiments on the test suite to evaluate commonsense reasoning in neural machine translation and investigate factors that have impact on this capability. Our experiments and analyses demonstrate that neural machine translation performs poorly on commonsense reasoning of the three ambiguity types in terms of both reasoning accuracy ( 6 60.1%) and reasoning consistency (6 31%). We will release our test suite as a machine translation commonsense reasoning testbed to promote future work in this direction.
Recent work proposes a family of contextual embeddings that significantly improves the accuracy of sequence labelers over non-contextual embeddings. However, there is no definite conclusion on whether we can build better sequence labelers by combining different kinds of embeddings in various settings. In this paper, we conduct extensive experiments on 3 tasks over 18 datasets and 8 languages to study the accuracy of sequence labeling with various embedding concatenations and make three observations: (1) concatenating more embedding variants leads to better accuracy in rich-resource and cross-domain settings and some conditions of low-resource settings; (2) concatenating contextual sub-word embeddings with contextual character embeddings hurts the accuracy in extremely low-resource settings; (3) based on the conclusion of (1), concatenating additional similar contextual embeddings cannot lead to further improvements. We hope these conclusions can help people build stronger sequence labelers in various settings.
Meta-embedding learning, which combines complementary information in different word embeddings, have shown superior performances across different Natural Language Processing tasks. However, domain-specific knowledge is still ignored by existing meta-embedding methods, which results in unstable performances across specific domains. Moreover, the importance of general and domain word embeddings is related to downstream tasks, how to regularize meta-embedding to adapt downstream tasks is an unsolved problem. In this paper, we propose a method to incorporate both domain-specific and task-oriented information into meta-embeddings. We conducted extensive experiments on four text classification datasets and the results show the effectiveness of our proposed method.
The linear-chain Conditional Random Field (CRF) model is one of the most widely-used neural sequence labeling approaches. Exact probabilistic inference algorithms such as the forward-backward and Viterbi algorithms are typically applied in training and prediction stages of the CRF model. However, these algorithms require sequential computation that makes parallelization impossible. In this paper, we propose to employ a parallelizable approximate variational inference algorithm for the CRF model. Based on this algorithm, we design an approximate inference network that can be connected with the encoder of the neural CRF model to form an end-to-end network, which is amenable to parallelization for faster training and prediction. The empirical results show that our proposed approaches achieve a 12.7-fold improvement in decoding speed with long sentences and a competitive accuracy compared with the traditional CRF approach.