Traditional click-through rate (CTR) prediction models convert the tabular data into one-hot vectors and leverage the collaborative relations among features for inferring the user’s preference over items. This modeling paradigm discards essential semantic information. Though some works like P5 and KAR have explored the potential of using Pre-trained Language Models (PLMs) to extract semantic signals for CTR prediction, they are computationally expensive and suffer from low efficiency. Besides, the beneficial collaborative relations are not considered, hindering the recommendation performance. To solve these problems, in this article, we propose a novel framework CTRL , which is industrial-friendly and model-agnostic with superior inference efficiency. Specifically, the original tabular data is first converted into textual data. Both tabular data and converted textual data are regarded as two different modalities and are separately fed into the collaborative CTR model and PLM. A cross-modal knowledge alignment procedure is performed to fine-grained align and integrate the collaborative and semantic signals, and the lightweight collaborative model can be deployed online for efficient serving after fine-tuning with supervised signals. Experimental results on three public datasets show that CTRL outperforms the state-of-the-art (SOTA) CTR models significantly. Moreover, we further verify its effectiveness on a large-scale industrial recommender system.

GPT-4o, an omni-modal model that enables vocal conversations with diverse emotions and tones, marks a milestone for omni-modal foundation models. However, empowering Large Language Models to perceive and generate images, texts, and speeches end-to-end with publicly available data remains challenging for the open-source community. Existing vision-language models rely on external tools for speech processing, while speech-language models still suffer from limited or totally without vision-understanding capabilities. To address this gap, we propose the EMOVA (EMotionally Omni-present Voice Assistant), to enable Large Language Models with end-to-end speech abilities while maintaining the leading vision-language performance. With a semantic-acoustic disentangled speech tokenizer, we surprisingly notice that omni-modal alignment can further enhance vision-language and speech abilities compared with the bi-modal aligned counterparts. Moreover, a lightweight style module is introduced for the flexible speech style controls including emotions and pitches. For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks, and meanwhile, supporting omni-modal spoken dialogue with vivid emotions.

High-resolution image inputs allow Large Vision-Language Models (LVLMs) to capture finer visual details, improving comprehension. However, the increased training and computational costs associated with such inputs pose significant challenges. A common approach to mitigate these costs involves slicing the input into uniform patches using sliding windows, each aligned with the vision encoder’s input size. While efficient, this method fragments the input, disrupting the continuity of context, which negatively impacts cross-patch perception tasks. To address these limitations, we propose HiRes-LLaVA, a novel framework designed to efficiently process high-resolution inputs of any size without altering the original contextual and geometric information. HiRes-LLaVA introduces two key components: (i) a SliceRestore Adapter (SRA) that reconstructs sliced patches into their original form, enabling efficient extraction of both global and local features through down-up-sampling and convolutional layers, and (ii) a Self-Mining Sampler (SMS) that compresses visual tokens based on internal relationships, preserving original context and positional information while reducing training overhead. To assess the ability of handling context fragmentation, we construct a new benchmark, EntityGrid-QA, consisting of edge-related tasks. Extensive experiments demonstrate the superiority of HiRes-LLaVA on both existing public benchmarks and EntityGrid-QA. For example, with SRA, our method achieves a performance improvement of ∼ 12% over state-of-the-art LVLMs in addressing fragmentation issues. Additionally, our SMS outperforms other visual token downsamplers, while offering high data efficiency.

In this paper, we introduce ILLUME, a unified multimodal large language model (MLLM) that seamlessly integrates multimodal understanding and generation capabilities within a single large language model through a unified next-token prediction formulation. To address the large dataset size typically required for image-text alignment, we propose to enhance data efficiency through the design of a vision tokenizer that incorporates semantic information and a progressive multi-stage training procedure. This approach reduces the dataset size to just 15M for pretraining -- over four times fewer than what is typically needed -- while achieving competitive or even superior performance with existing unified MLLMs, such as Janus. Additionally, to promote synergistic enhancement between understanding and generation capabilities, which is under-explored in previous works, we introduce a novel self-enhancing multimodal alignment scheme. This scheme supervises the MLLM to self-assess the consistency between text descriptions and self-generated images, facilitating the model to interpret images more accurately and avoid unrealistic and incorrect predictions caused by misalignment in image generation. Based on extensive experiments, our proposed ILLUME stands out and competes with state-of-the-art unified MLLMs and specialized models across various benchmarks for multimodal understanding, generation, and editing.

Recently, quantization has been widely used for the compression and acceleration of large language models (LLMs). Due to the outliers in LLMs, it is crucial to flatten weights and activations to minimize quantization error with equally spaced quantization points. Prior research explores various pre-quantization transformations to suppress outliers, such as per-channel scaling and Hadamard transformation. However, we observe that these transformed weights and activations can still exhibit steep and dispersed distributions. In this paper, we propose FlatQuant (Fast and Learnable Affine Transformation), a new post-training quantization approach that enhances the flatness of weights and activations. Our approach identifies optimal affine transformations for each linear layer, calibrated in hours via a lightweight objective. To reduce runtime overhead of affine transformation, we apply Kronecker product with two lightweight matrices, and fuse all operations in FlatQuant into a single kernel. Extensive experiments demonstrate that FlatQuant establishes a new state-of-the-art benchmark for quantization. For example, it achieves less than 1\% accuracy drop for W4A4 quantization on the LLaMA-3-70B model, surpassing SpinQuant by 7.5\%. Additionally, it provides up to 2.3x prefill speedup and 1.7x decoding speedup compared to the FP16 model. Code is available at: https://github.com/ruikangliu/FlatQuant.

Layer pruning has emerged as a widely used technique for compressing large language models (LLMs). However, existing layer pruning approaches often incur substantial performance degradation. We identify the majority of this degradation to a single yet previously overlooked issue: \textit{the mismatch of activation magnitudes at the pruning interface}. The pre-interface activations exhibit significantly different scales from the post-interface ones, causing the distributional shift as it propagates through the remaining layers. To address this issue, we introduce \textsc{LinearPatch}, a lightweight and plug-and-play technique that fuses two operations into one matrix multiply at the pruning interface: (i) a Hadamard transformation that suppresses massive outliers at particular tokens and (ii) a channel-wise scaling that aligns activation statistics. On LLaMA-3-8B, \textsc{LinearPatch} preserves up to \textbf{94.15\%} of the original model's performance when pruning 5 out of 32 layers, outperforming the previous state of the art by \textbf{4\%}. The patch can be further refined with 5K unlabeled samples via memory-efficient offline distillation, pushing the retention to 95.16\% within only 30 minutes on a single GPU. Code is available at \url{https://github.com/chenxinrui-tsinghua/LinearPatch}.

Vision-language pretrained models have achieved impressive performance on various downstream tasks. However, their large model sizes hinder their utilization on platforms with limited computational resources. We find that directly using smaller pretrained models and applying magnitude-based pruning on CLIP models leads to in-flexibility and inferior performance. Recent efforts for VLP compression either adopt uni-modal compression metrics resulting in limited performance or involve costly mask-search processes with learnable masks. In this paper, we first propose the Module-wise Pruning Error (MoPE) met-ric, accurately assessing CLIP module importance by performance decline on cross-modal tasks. Using the MoPE metric, we introduce a unified pruning framework applica-ble to both pretraining and task-specific fine-tuning compression stages. For pretraining, MoPE-CLIP effectively leverages knowledge from the teacher model, significantly reducing pretraining costs while maintaining strong zero-shot capabilities. For fine-tuning, consecutive pruning from width to depth yields highly competitive task-specific models. Extensive experiments in two stages demonstrate the effectiveness of the MoPE metric, and MoPE-CLIP outperforms previous state-of-the-art VLP compression methods.

This work proposes TimeChat, a time-sensitive multi-modal large language model specifically designed for long video understanding. Our model incorporates two key architectural contributions: (1) a timestamp-aware frame encoder that binds visual content with the timestamp of each frame, and (2) a sliding video Q-Former that produces a video token sequence of varying lengths to accommodate videos of various durations. Additionally, we construct an instruction-tuning dataset, encompassing 6 tasks and a total of 125K instances, to further enhance TimeChat's instruction-following performance. Experiment results across various video understanding tasks, such as dense captioning, temporal grounding, and highlight detection, demonstrate TimeChat's strong zero-shot temporal localization and reasoning capabilities. For example, it achieves +9.2 F1 score and +2.8 CIDEr on YouCook2, +5.8 HIT@1 on QVHighlights, and +27.5 R@1 (I oU=0.5) on Charades-STA, compared to state-of-the-art video large language models, holding the potential to serve as a versatile video assistant for long-form video comprehension tasks and satisfy realistic user requirements.11Our code and dataset are available at https://github.com/RenShuhuai-Andy/TimeChat.

With the rapid growth of large language models (LLMs), there is increasing demand for memory and computation in LLMs. Recent efforts on post-training pruning of LLMs aim to reduce the model size and computation requirements, yet the performance is still sub-optimal. In this paper, we present a plug-and-play solution for post-training pruning of LLMs. The proposed solution has two innovative components: 1) **Relative Importance and Activations (RIA)**, a new pruning metric that jointly considers the weight and activations efficiently on LLMs, and 2) **Channel Permutation**, a new approach to maximally preserves important weights under N:M sparsity. The two proposed components can be readily combined to further enhance the N:M semi-structured pruning of LLMs. Our empirical experiments show that RIA alone can already surpass all existing post-training pruning methods on prevalent LLMs, e.g., LLaMA ranging from 7B to 65B. Furthermore, N:M semi-structured pruning with channel permutation can even outperform the original LLaMA2-70B on zero-shot tasks, together with practical speed-up on specific hardware. Our code is available at: https://github.com/biomedical-cybernetics/Relative-importance-and-activation-pruning

Zhiming Mao, Haoli Bai, Lu Hou, Lifeng Shang, Xin Jiang, Qun Liu, Kam-Fai Wong. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Currently, vision encoder models like Vision Transformers (ViTs) typically excel at image recognition tasks but cannot simultaneously support text recognition like human visual recognition. To address this limitation, we propose UNIT, a novel training framework aimed at UNifying Image and Text recognition within a single model. Starting with a vision encoder pre-trained with image recognition tasks, UNIT introduces a lightweight language decoder for predicting text outputs and a lightweight vision decoder to prevent catastrophic forgetting of the original image encoding capabilities. The training process comprises two stages: intra-scale pretraining and inter-scale finetuning. During intra-scale pretraining, UNIT learns unified representations from multi-scale inputs, where images and documents are at their commonly used resolution, to enable fundamental recognition capability. In the inter-scale finetuning stage, the model introduces scale-exchanged data, featuring images and documents at resolutions different from the most commonly used ones, to enhance its scale robustness. Notably, UNIT retains the original vision encoder architecture, making it cost-free in terms of inference and deployment. Experiments across multiple benchmarks confirm that our method significantly outperforms existing methods on document-related tasks (e.g., OCR and DocQA) while maintaining the performances on natural images, demonstrating its ability to substantially enhance text recognition without compromising its core image recognition capabilities.

Training deep neural networks (DNNs) typically requires massive computational power. Existing DNNs exhibit low time and storage efficiency due to the high degree of redundancy. In contrast to most existing DNNs, biological and social networks with vast numbers of connections are highly efficient and exhibit scale-free properties indicative of the power law distribution, which can be originated by preferential attachment in growing networks. In this work, we ask whether the topology of the best performing DNNs shows the power law similar to biological and social networks and how to use the power law topology to construct well-performing and compact DNNs. We first find that the connectivities of sparse DNNs can be modeled by truncated power law distribution, which is one of the variations of the power law. The comparison of different DNNs reveals that the best performing networks correlated highly with the power law distribution. We further model the preferential attachment in DNNs evolution and find that continual learning in networks with growth in tasks correlates with the process of preferential attachment. These identified power law dynamics in DNNs can lead to the construction of highly accurate and compact DNNs based on preferential attachment. Inspired by the discovered findings, two novel applications have been proposed, including evolving optimal DNNs in sparse network generation and continual learning tasks with efficient network growth using power law dynamics. Experimental results indicate that the proposed applications can speed up training, save storage, and learn with fewer samples than other well-established baselines. Our demonstration of preferential attachment and power law in well-performing DNNs offers insight into designing and constructing more efficient deep learning.

Embedding tables are usually huge in click-through rate (CTR) prediction models. To train and deploy the CTR models efficiently and economically, it is necessary to compress their embedding tables. To this end, we formulate a novel quantization training paradigm to compress the embeddings from the training stage, termed low-precision training (LPT). Also, we provide theoretical analysis on its convergence. The results show that stochastic weight quantization has a faster convergence rate and a smaller convergence error than deterministic weight quantization in LPT. Further, to reduce accuracy degradation, we propose adaptive low-precision training (ALPT) which learns the step size (i.e., the quantization resolution). Experiments on two real-world datasets confirm our analysis and show that ALPT can significantly improve the prediction accuracy, especially at extremely low bit width. For the first time in CTR models, we successfully train 8-bit embeddings without sacrificing prediction accuracy.

Haoli Bai, Zhiguang Liu, Xiaojun Meng, Li Wentao, Shuang Liu, Yifeng Luo, Nian Xie, Rongfu Zheng, Liangwei Wang, Lu Hou, Jiansheng Wei, Xin Jiang, Qun Liu. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Guanhua Chen, Lu Hou, Yun Chen, Wenliang Dai, Lifeng Shang, Xin Jiang, Qun Liu, Jia Pan, Wenping Wang. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Recently, open-domain text-to-video (T2V) generation models have made remarkable progress. However, the promising results are mainly shown by the qualitative cases of generated videos, while the quantitative evaluation of T2V models still faces two critical problems. Firstly, existing studies lack fine-grained evaluation of T2V models on different categories of text prompts. Although some benchmarks have categorized the prompts, their categorization either only focuses on a single aspect or fails to consider the temporal information in video generation. Secondly, it is unclear whether the automatic evaluation metrics are consistent with human standards. To address these problems, we propose FETV, a benchmark for Fine-grained Evaluation of Text-to-Video generation. FETV is multi-aspect, categorizing the prompts based on three orthogonal aspects: the major content, the attributes to control and the prompt complexity. FETV is also temporal-aware, which introduces several temporal categories tailored for video generation. Based on FETV, we conduct comprehensive manual evaluations of four representative T2V models, revealing their pros and cons on different categories of prompts from different aspects. We also extend FETV as a testbed to evaluate the reliability of automatic T2V metrics. The multi-aspect categorization of FETV enables fine-grained analysis of the metrics' reliability in different scenarios. We find that existing automatic metrics (e.g., CLIPScore and FVD) correlate poorly with human evaluation. To address this problem, we explore several solutions to improve CLIPScore and FVD, and develop two automatic metrics that exhibit significant higher correlation with humans than existing metrics. Benchmark page: https://github.com/llyx97/FETV.

Chaofan Tao, Lu Hou, Wei Zhang, Lifeng Shang, Xin Jiang, Qun Liu, Ping Luo, Ngai Wong. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Recent large-scale video-language pre-trained models have shown appealing performance on various downstream tasks.However, the pretraining process is computationally expensive due to the requirement of millions of videotext pairs and the redundant data structure of each video.To mitigate these problems, we propose LiteVL, which adapts a pre-trained image-language model BLIP into a video-text model directly on downstream tasks, without heavy pre-training.To enhance the temporal modeling lacking in the image-language model, we propose to add temporal attention modules in the image encoder of BLIP with dynamic temporal scaling.Besides the model-wise adaptation, we also propose a non-parametric pooling mechanism to adaptively reweight the fine-grained video embedding conditioned on the text.Experimental results on text-video retrieval and video question answering show that the proposed LiteVL even outperforms previous video-language pre-trained models by a clear margin, though without any videolanguage pre-training.

Network quantization has gained increasing attention with the rapid growth of large pre-trained language models~(PLMs). However, most existing quantization methods for PLMs follow quantization-aware training~(QAT) that requires end-to-end training with full access to the entire dataset. Therefore, they suffer from slow training, large memory overhead, and data accessibility issues. In this paper, we study post-training quantization~(PTQ) of PLMs, and propose module-wise quantization error minimization~(MREM), an efficient solution to mitigate these issues. By partitioning the PLM into multiple modules, we minimize the reconstruction error incurred by quantization for each module. In addition, we design a new model parallel training strategy such that each module can be trained locally on separate computing devices without waiting for preceding modules, which brings nearly the theoretical training speed-up (e.g., $4\times$ on $4$ GPUs). Experiments on GLUE and SQuAD benchmarks show that our proposed PTQ solution not only performs close to QAT, but also enjoys significant reductions in training time, memory overhead, and data consumption.

Vision-Language Pre-training (VLP) models have shown remarkable performance on various downstream tasks. Their success heavily relies on the scale of pre-trained cross-modal datasets. However, the lack of large-scale datasets and benchmarks in Chinese hinders the development of Chinese VLP models and broader multilingual applications. In this work, we release a large-scale Chinese cross-modal dataset named Wukong, which contains 100 million Chinese image-text pairs collected from the web. Wukong aims to benchmark different multi-modal pre-training methods to facilitate the VLP research and community development. Furthermore, we release a group of models pre-trained with various image encoders (ViT-B/ViT-L/SwinT) and also apply advanced pre-training techniques into VLP such as locked-image text tuning, token-wise similarity in contrastive learning, and reduced-token interaction. Extensive experiments and a benchmarking of different downstream tasks including a new largest human-verified image-text test dataset are also provided. Experiments show that Wukong can serve as a promising Chinese pre-training dataset and benchmark for different cross-modal learning methods. For the zero-shot image classification task on 10 datasets, $Wukong_\text{ViT-L}$ achieves an average accuracy of 73.03%. For the image-text retrieval task, it achieves a mean recall of 71.6% on AIC-ICC which is 12.9% higher than WenLan 2.0. Also, our Wukong models are benchmarked on downstream tasks with other variants on multiple datasets, e.g., Flickr8K-CN, Flickr-30K-CN, COCO-CN, et al. More information can be referred to https://wukong-dataset.github.io/wukong-dataset/.

Haoli Bai, Wei Zhang, Lu Hou, Lifeng Shang, Jin Jin, Xin Jiang, Qun Liu, Michael Lyu, Irwin King. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Zhiqi Huang, Lu Hou, Lifeng Shang, Xin Jiang, Xiao Chen, Qun Liu. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Recently, learning a model that generalizes well on out-of-distribution (OOD) data has attracted great attention in the machine learning community. In this paper, after defining OOD generalization by Wasserstein distance, we theoretically justify that a model robust to input perturbation also generalizes well on OOD data. Inspired by previous findings that adversarial training helps improve robustness, we show that models trained by adversarial training have converged excess risk on OOD data. Besides, in the paradigm of pre-training then fine-tuning, we theoretically justify that the input perturbation robust model in the pre-training stage provides an initialization that generalizes well on downstream OOD data. Finally, various experiments conducted on image classification and natural language understanding tasks verify our theoretical findings.

Transformer-based pre-training models like BERT have achieved remarkable performance in many natural language processing tasks.However, these models are both computation and memory expensive, hindering their deployment to resource-constrained devices.In this work, we propose TernaryBERT, which ternarizes the weights in a fine-tuned BERT model.Specifically, we use both approximation-based and loss-aware ternarization methods and empirically investigate the ternarization granularity of different parts of BERT.Moreover, to reduce the accuracy degradation caused by the lower capacity of low bits, we leverage the knowledge distillation technique (Jiao et al., 2019) in the training process.Experiments on the GLUE benchmark and SQuAD show that our proposed TernaryBERT outperforms the other BERT quantization methods, and even achieves comparable performance as the fullprecision model while being 14.9x smaller.

The pre-trained language models like BERT, though powerful in many natural language processing tasks, are both computation and memory expensive. To alleviate this problem, one approach is to compress them for specific tasks before deployment. However, recent works on BERT compression usually compress the large BERT model to a fixed smaller size, and can not fully satisfy the requirements of different edge devices with various hardware performances. In this paper, we propose a novel dynamic BERT model (abbreviated as DynaBERT), which can flexibly adjust the size and latency by selecting adaptive width and depth. The training process of DynaBERT includes first training a width-adaptive BERT and then allowing both adaptive width and depth, by distilling knowledge from the full-sized model to small sub-networks. Network rewiring is also used to keep the more important attention heads and neurons shared by more sub-networks. Comprehensive experiments under various efficiency constraints demonstrate that our proposed dynamic BERT (or RoBERTa) at its largest size has comparable performance as BERT-base (or RoBERTa-base), while at smaller widths and depths consistently outperforms existing BERT compression methods. Code is available at https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/DynaBERT.

The long-short-term memory (LSTM), though powerful, is memory and computa\x02tion expensive. To alleviate this problem, one approach is to compress its weights by quantization. However, existing quantization methods usually have inferior performance when used on LSTMs. In this paper, we first show theoretically that training a quantized LSTM is difficult because quantization makes the exploding gradient problem more severe, particularly when the LSTM weight matrices are large. We then show that the popularly used weight/layer/batch normalization schemes can help stabilize the gradient magnitude in training quantized LSTMs. Empirical results show that the normalized quantized LSTMs achieve significantly better results than their unnormalized counterparts. Their performance is also comparable with the full-precision LSTM, while being much smaller in size.

In timeseries classification, shapelets are subsequences of timeseries with high discriminative power. Existing methods perform a combinatorial search for shapelet discovery. Even with speedup heuristics such as pruning, clustering, and dimensionality reduction, the search remains computationally expensive. In this paper, we take an entirely different approach and reformulate the shapelet discovery task as a numerical optimization problem. In particular, the shapelet positions are learned by combining the generalized eigenvector method and fused lasso regularizer to encourage a sparse and blocky solution. Extensive experimental results show that the proposed method is orders of magnitudes faster than the state-of-the-art shapelet-based methods, while achieving comparable or even better classification accuracy.