Handling long-context sequences efficiently remains a significant challenge in large language models (LLMs). Existing methods for token selection in sequence extrapolation either employ a permanent eviction strategy or select tokens by chunk, which may lead to the loss of critical information. We propose Efficient Selective Attention (ESA), a novel approach that extends context length by efficiently selecting the most critical tokens at the token level to compute attention. ESA reduces the computational complexity of token selection by compressing query and key vectors into lower-dimensional representations. We evaluate ESA on long sequence benchmarks with maximum lengths up to 256k using open-source LLMs with context lengths of 8k and 32k. ESA outperforms other selective attention methods, especially in tasks requiring the retrieval of multiple pieces of information, achieving comparable performance to full-attention extrapolation methods across various tasks, with superior results in certain tasks.

Diffusion models have shown exceptional performance in visual generation tasks. Recently, these models have shifted from traditional U-Shaped CNN-Attention hybrid structures to fully transformer-based isotropic architectures. While these transformers exhibit strong scalability and performance, their reliance on complicated self-attention operation results in slow inference speeds. Contrary to these works, we rethink one of the simplest yet fastest module in deep learning, 3x3 Convolution, to construct a scaled-up purely convolutional diffusion model. We first discover that an Encoder-Decoder Hourglass design outperforms scalable isotropic architectures for Conv3x3, but still under-performing our expectation. Further improving the architecture, we introduce sparse skip connections to reduce redundancy and improve scalability. Based on the architecture, we introduce conditioning improvements including stage-specific embeddings, mid-block condition injection, and conditional gating. These improvements lead to our proposed Diffusion CNN (DiC), which serves as a swift yet competitive diffusion architecture baseline. Experiments on various scales and settings show that DiC surpasses existing diffusion transformers by considerable margins in terms of performance while keeping a good speed advantage. Project page: https://github.com/YuchuanTian/DiC

Knowledge distillation plays a key role in compressing the Large Language Models (LLMs), which boosts a small-size student model under large teacher models' guidance. However, existing LLM distillation methods overly rely on student-generated outputs, which may introduce generation errors and misguide the distillation process. Moreover, the distillation loss functions introduced in previous art struggle to align the most informative part due to the complex distribution of LLMs' outputs. To address these problems, we propose a multi-granularity semantic revision method for LLM distillation. At the sequence level, we propose a sequence correction and re-generation (SCRG) strategy. SCRG first calculates the semantic cognitive difference between the teacher and student to detect the error token, then corrects it with the teacher-generated one, and re-generates the sequence to reduce generation errors and enhance generation diversity. At the token level, we design a distribution adaptive clipping Kullback-Leibler (DAC-KL) loss as the distillation objective function. DAC-KL loss exploits a learnable sub-network to adaptively extract semantically dense areas from the teacher's output, avoiding the interference of redundant information in the distillation process. Finally, at the span level, we leverage the span priors of a sequence to compute the probability correlations within spans, and constrain the teacher and student's probability correlations to be consistent, further enhancing the transfer of semantic information. Extensive experiments across different model families with parameters ranging from 0.1B to 13B demonstrate the superiority of our method compared to existing methods.

Recent semi-supervised object detection (SSOD) has achieved remarkable progress by leveraging unlabeled data for training. Mainstream SSOD methods rely on Consistency Regularization methods and Exponential Moving Average (EMA), which form a cyclic data flow. However, the EMA updating training approach leads to weight coupling between the teacher and student models. This coupling in a cyclic data flow results in a decrease in the utilization of unlabeled data information and the confirmation bias on low-quality or erroneous pseudo-labels. To address these issues, we propose the Collaboration of Teachers Framework (CTF), which consists of multiple pairs of teacher and student models for training. In the learning process of CTF, the Data Performance Consistency Optimization module (DPCO) informs the best pair of teacher models possessing the optimal pseudo-labels during the past training process, and these most reliable pseudo-labels generated by the best performing teacher would guide the other student models. As a consequence, this framework greatly improves the utilization of unlabeled data and prevents the positive feedback cycle of unreliable pseudo-labels. The CTF achieves outstanding results on numerous SSOD datasets, including a 0.71% mAP improvement on the 10% annotated COCO dataset and a 0.89% mAP improvement on the VOC dataset compared to LabelMatch and converges significantly faster. Moreover, the CTF is plug-and-play and can be integrated with other mainstream SSOD methods.

In an effort to reduce the computational load of Transformers, research on linear attention has gained significant momentum. However, the improvement strategies for attention mechanisms typically necessitate extensive retraining, which is impractical for large language models with a vast array of parameters. In this paper, we present DiJiang, a novel Frequency Domain Kernelization approach that enables the transformation of a pre-trained vanilla Transformer into a linear complexity model with little training costs. By employing a weighted Quasi-Monte Carlo method for sampling, the proposed approach theoretically offers superior approximation efficiency. To further reduce the training computational complexity, our kernelization is based on Discrete Cosine Transform (DCT) operations. Extensive experiments demonstrate that the proposed method achieves comparable performance to the original Transformer, but with significantly reduced training costs and much faster inference speeds. Our DiJiang-7B achieves comparable performance with LLaMA2-7B on various benchmark while requires only about 1/50 training cost. Code is available at https://github.com/YuchuanTian/DiJiang.

Post-training quantization (PTQ) has played a key role in compressing large language models (LLMs) with ultra-low costs. However, existing PTQ methods only focus on handling the outliers within one layer or one block, which ignores the dependency of blocks and leads to severe performance degradation in low-bit settings. In this paper, we propose CBQ, a cross-block reconstruction-based PTQ method for LLMs. CBQ employs a cross-block dependency using a homologous reconstruction scheme, establishing long-range dependencies across multiple blocks to minimize error accumulation. Furthermore, CBQ incorporates a coarse-to-fine preprocessing (CFP) strategy for suppressing weight and activation outliers, coupled with an adaptive LoRA-Rounding technique for precise weight quantization. These innovations enable CBQ to not only handle extreme outliers effectively but also improve overall quantization accuracy. Extensive experiments show that CBQ achieves superior low-bit quantization (W4A4, W4A8, W2A16) and outperforms existing state-of-the-art methods across various LLMs and datasets. Notably, CBQ quantizes the 4-bit LLAMA1-65B model within only 4.3 hours on a single GPU, achieving a commendable tradeoff between performance and quantization efficiency.

Abstract--The recent trend of large language models (LLMs) is to increase the scale of both model size (a.k.a the number of parameters) and dataset to achieve better generative ability, which is definitely proved by a lot of work such as the famous GPT and Llama. However, large models often involve massive computational costs, and practical applications cannot afford such high prices. However, the method of constructing a strong model architecture for LLMs is rarely discussed. We first analyze the state-of-the-art language model architectures and observe the feature collapse problem. Based on the theoretical analysis, we propose that the nonlinearity is also very important for language models, which is usually studied in convolutional neural networks for vision tasks. The series informed activation function is then introduced with tiny calculations that can be ignored, and an augmented shortcut is further used to enhance the model nonlinearity. We then demonstrate that the proposed approach is significantly effective for enhancing the model nonlinearity through carefully designed ablations; thus, we present a new efficient model architecture for establishing modern, namely, PanGu- π . Experiments are then conducted using the same dataset and training strategy to compare PanGu- π with state-of-the-art LLMs. The results show that PanGu- π -7B can achieve a comparable performance to that of benchmarks with about 10% inference speed-up, and PanGu- π -1B can achieve state-of-the-art performance in terms of accuracy and efficiency. In addition, we have deployed PanGu- π -7B in the high-value domains of finance and law, developing an LLM named YunShan for practical application. The results show that YunShan can surpass other models with similar scales on benchmarks. As shown in Figure 1, our translation, text summarization, and dialogue.

Knowledge distillation (KD) emerges as a promising yet challenging technique for compressing deep neural networks, aiming to transfer extensive learning representations from proficient and computationally intensive teacher models to compact student models. However, current KD methods for super-resolution (SR) models have limited performance and restricted applications, since the characteristics of SR tasks are overlooked. In this paper, we put forth an approach from the perspective of effective data utilization, namely, the Data Upcycling Knowledge Distillation (DUKD), which facilitates the student model by the prior knowledge the teacher provided through the upcycled in-domain data derived from the input images. Besides, for the first time, we realize the label consistency regularization in KD for SR models, which is implemented by the paired invertible data augmentations. It constrains the training process of KD and leads to better generalization capability of the student model. The DUKD, due to its versatility, can be applied across a broad spectrum of teacher-student architectures (e.g., CNN and Transformer models) and SR tasks, such as single image SR, real-world SR, and SR quantization, and is in parallel with other compression techniques. Comprehensive experiments on diverse benchmarks demonstrate that the DUKD method significantly outperforms previous art.

Recent releases of Large Language Models (LLMs), e.g. ChatGPT, are astonishing at generating human-like texts, but they may impact the authenticity of texts. Previous works proposed methods to detect these AI-generated texts, including simple ML classifiers, pretrained-model-based zero-shot methods, and finetuned language classification models. However, mainstream detectors always fail on short texts, like SMSes, Tweets, and reviews. In this paper, a Multiscale Positive-Unlabeled (MPU) training framework is proposed to address the difficulty of short-text detection without sacrificing long-texts. Firstly, we acknowledge the human-resemblance property of short machine texts, and rephrase AI text detection as a partial Positive-Unlabeled (PU) problem by regarding these short machine texts as partially "unlabeled". Then in this PU context, we propose the length-sensitive Multiscale PU Loss, where a recurrent model in abstraction is used to estimate positive priors of scale-variant corpora. Additionally, we introduce a Text Multiscaling module to enrich training corpora. Experiments show that our MPU method augments detection performance on long AI-generated texts, and significantly improves short-text detection of language model detectors. Language Models trained with MPU could outcompete existing detectors on various short-text and long-text detection benchmarks. The codes are available at https://github.com/mindspore-lab/mindone/tree/master/examples/detect_chatgpt and https://github.com/YuchuanTian/AIGC_text_detector.

Transformer is a type of deep neural network mainly based on self-attention mechanism which is originally applied in natural language processing field. Inspired by the strong representation ability of transformer, researchers propose to extend transformer for computer vision tasks. Transformer-based models show competitive and even better performance on various visual benchmarks compared to other network types such as convolutional networks and recurrent networks. With high performance and without inductive bias defined by human, transformer is receiving more and more attention from the visual community. In this paper we provide a literature review of these visual transformer models by categorizing them in different tasks and analyze the advantages and disadvantages of these methods. In particular, the main categories include the basic image classification, high-level vision, low-level vision and video processing. The self-attention in computer vision is also briefly revisited as self-attention is the base component in transformer. Efficient transformer methods are included for pushing transformer into real applications on the devices. Finally, we give a discussion about the challenges and further research directions for visual transformers.