Numerical Pruning for Efficient Autoregressive Models

Transformers have emerged as the leading architecture in deep learning, proving to be versatile and highly effective across diverse domains beyond language and image processing. However, their impressive performance often incurs high computational costs due to their substantial model size. This paper focuses on compressing decoder-only transformer-based autoregressive models through structural weight pruning to improve the model efficiency while preserving performance for both language and image generation tasks. Specifically, we propose a training-free pruning method that calculates a numerical score with Newton's method for the Attention and MLP modules, respectively. Besides, we further propose another compensation algorithm to recover the pruned model for better performance. To verify the effectiveness of our method, we provide both theoretical support and extensive experiments. Our experiments show that our method achieves state-of-the-art performance with reduced memory usage and faster generation speeds on GPUs.

LazyDiT: Lazy Learning for the Acceleration of Diffusion Transformers

Diffusion Transformers have emerged as the preeminent models for a wide array of generative tasks, demonstrating superior performance and efficacy across various applications. The promising results come at the cost of slow inference, as each denoising step requires running the whole transformer model with a large amount of parameters. In this paper, we show that performing the full computation of the model at each diffusion step is unnecessary, as some computations can be skipped by lazily reusing the results of previous steps. Furthermore, we show that the lower bound of similarity between outputs at consecutive steps is notably high, and this similarity can be linearly approximated using the inputs. To verify our demonstrations, we propose the \textbf{LazyDiT}, a lazy learning framework that efficiently leverages cached results from earlier steps to skip redundant computations. Specifically, we incorporate lazy learning layers into the model, effectively trained to maximize laziness, enabling dynamic skipping of redundant computations. Experimental results show that LazyDiT outperforms the DDIM sampler across multiple diffusion transformer models at various resolutions. Furthermore, we implement our method on mobile devices, achieving better performance than DDIM with similar latency.

MIND: Effective Incorrect Assignment Detection through a Multi-Modal Structure-Enhanced Language Model

The rapid growth of academic publications has exacerbated the issue of author name ambiguity in online digital libraries. Despite advances in name disambiguation algorithms, cumulative errors continue to undermine the reliability of academic systems. It is estimated that over 10% paper-author assignments are rectified when constructing the million-scale WhoIsWho benchmark. Existing endeavors to detect incorrect assignments are either semantic-based or graph-based approaches, which fall short of making full use of the rich text attributes of papers and implicit structural features defined via the co-occurrence of paper attributes. To this end, this paper introduces a structure-enhanced language model that combines key structural features from graph-based methods with fine-grained semantic features from rich paper attributes to detect incorrect assignments. The proposed model is trained with a highly effective multi-modal multi-turn instruction tuning framework, which incorporates task-guided instruction tuning, text-attribute modality, and structural modality. Experimental results demonstrate that our model outperforms previous approaches, achieving top performance on the leaderboard of KDD Cup 2024. Our code has been publicly available.

LIBER: Lifelong User Behavior Modeling Based on Large Language Models

CTR prediction plays a vital role in recommender systems. Recently, large language models (LLMs) have been applied in recommender systems due to their emergence abilities. While leveraging semantic information from LLMs has shown some improvements in the performance of recommender systems, two notable limitations persist in these studies. First, LLM-enhanced recommender systems encounter challenges in extracting valuable information from lifelong user behavior sequences within textual contexts for recommendation tasks. Second, the inherent variability in human behaviors leads to a constant stream of new behaviors and irregularly fluctuating user interests. This characteristic imposes two significant challenges on existing models. On the one hand, it presents difficulties for LLMs in effectively capturing the dynamic shifts in user interests within these sequences, and on the other hand, there exists the issue of substantial computational overhead if the LLMs necessitate recurrent calls upon each update to the user sequences. In this work, we propose Lifelong User Behavior Modeling (LIBER) based on large language models, which includes three modules: (1) User Behavior Streaming Partition (UBSP), (2) User Interest Learning (UIL), and (3) User Interest Fusion (UIF). Initially, UBSP is employed to condense lengthy user behavior sequences into shorter partitions in an incremental paradigm, facilitating more efficient processing. Subsequently, UIL leverages LLMs in a cascading way to infer insights from these partitions. Finally, UIF integrates the textual outputs generated by the aforementioned processes to construct a comprehensive representation, which can be incorporated by any recommendation model to enhance performance. LIBER has been deployed on Huawei's music recommendation service and achieved substantial improvements in users' play count and play time by 3.01% and 7.69%.

Circuit Complexity Bounds for RoPE-based Transformer Architecture

Characterizing the express power of the Transformer architecture is critical to understanding its capacity limits and scaling law. Recent works provide the circuit complexity bounds to Transformer-like architecture. On the other hand, Rotary Position Embedding ($\mathsf{RoPE}$) has emerged as a crucial technique in modern large language models, offering superior performance in capturing positional information compared to traditional position embeddings, which shows great potential in application prospects, particularly for the long context scenario. Empirical evidence also suggests that $\mathsf{RoPE}$-based Transformer architectures demonstrate greater generalization capabilities compared to conventional Transformer models. In this work, we establish a tighter circuit complexity bound for Transformers with $\mathsf{RoPE}$ attention. Our key contribution is that we show that unless $\mathsf{TC}^0 = \mathsf{NC}^1$, a $\mathsf{RoPE}$-based Transformer with $\mathrm{poly}(n)$-precision, $O(1)$ layers, hidden dimension $d \leq O(n)$ cannot solve the arithmetic problem or the Boolean formula value problem. This result significantly demonstrates the fundamental limitation of the expressivity of the $\mathsf{RoPE}$-based Transformer architecture, although it achieves giant empirical success. Our theoretical framework not only establishes tighter complexity bounds but also may instruct further work on the $\mathsf{RoPE}$-based Transformer.

Training Compute-Optimal Protein Language Models

We explore optimally training protein language models, an area of significant interest in biological research where guidance on best practices is limited. Most models are trained with extensive compute resources until performance gains plateau, focusing primarily on increasing model sizes rather than optimizing the efficient compute frontier that balances performance and compute budgets. Our investigation is grounded in a massive dataset consisting of 939 million protein sequences. We trained over 300 models ranging from 3.5 million to 10.7 billion parameters on 5 to 200 billion unique tokens, to investigate the relations between model sizes, training token numbers, and objectives. First, we observed the effect of diminishing returns for the Causal Language Model (CLM) and that of overfitting for the Masked Language Model~(MLM) when repeating the commonly used Uniref database. To address this, we included metagenomic protein sequences in the training set to increase the diversity and avoid the plateau or overfitting effects. Second, we obtained the scaling laws of CLM and MLM on Transformer, tailored to the specific characteristics of protein sequence data. Third, we observe a transfer scaling phenomenon from CLM to MLM, further demonstrating the effectiveness of transfer through scaling behaviors based on estimated Effectively Transferred Tokens. Finally, to validate our scaling laws, we compare the large-scale versions of ESM-2 and PROGEN2 on downstream tasks, encompassing evaluations of protein generation as well as structure- and function-related tasks, all within less or equivalent pre-training compute budgets.

Beyond Positive History: Re-ranking with List-level Hybrid Feedback

As the last stage of recommender systems, re-ranking generates a re-ordered list that aligns with the user's preference. However, previous works generally focus on item-level positive feedback as history (e.g., only clicked items) and ignore that users provide positive or negative feedback on items in the entire list. This list-level hybrid feedback can reveal users' holistic preferences and reflect users' comparison behavior patterns manifesting within a list. Such patterns could predict user behaviors on candidate lists, thus aiding better re-ranking. Despite appealing benefits, extracting and integrating preferences and behavior patterns from list-level hybrid feedback into re-ranking multiple items remains challenging. To this end, we propose Re-ranking with List-level Hybrid Feedback (dubbed RELIFE). It captures user's preferences and behavior patterns with three modules: a Disentangled Interest Miner to disentangle the user's preferences into interests and disinterests, a Sequential Preference Mixer to learn users' entangled preferences considering the context of feedback, and a Comparison-aware Pattern Extractor to capture user's behavior patterns within each list. Moreover, for better integration of patterns, contrastive learning is adopted to align the behavior patterns of candidate and historical lists. Extensive experiments show that RELIFE significantly outperforms SOTA re-ranking baselines.

Bypassing the Exponential Dependency: Looped Transformers Efficiently Learn In-context by Multi-step Gradient Descent

In-context learning has been recognized as a key factor in the success of Large Language Models (LLMs). It refers to the model's ability to learn patterns on the fly from provided in-context examples in the prompt during inference. Previous studies have demonstrated that the Transformer architecture used in LLMs can implement a single-step gradient descent update by processing in-context examples in a single forward pass. Recent work has further shown that, during in-context learning, a looped Transformer can implement multi-step gradient descent updates in forward passes. However, their theoretical results require an exponential number of in-context examples, $n = \exp(\Omega(T))$, where $T$ is the number of loops or passes, to achieve a reasonably low error. In this paper, we study linear looped Transformers in-context learning on linear vector generation tasks. We show that linear looped Transformers can implement multi-step gradient descent efficiently for in-context learning. Our results demonstrate that as long as the input data has a constant condition number, e.g., $n = O(d)$, the linear looped Transformers can achieve a small error by multi-step gradient descent during in-context learning. Furthermore, our preliminary experiments validate our theoretical analysis. Our findings reveal that the Transformer architecture possesses a stronger in-context learning capability than previously understood, offering new insights into the mechanisms behind LLMs and potentially guiding the better design of efficient inference algorithms for LLMs.

HSR-Enhanced Sparse Attention Acceleration

Large Language Models (LLMs) have demonstrated remarkable capabilities across various applications, but their performance on long-context tasks is often limited by the computational complexity of attention mechanisms. This paper introduces a novel approach to accelerate attention computation in LLMs, particularly for long-context scenarios. We leverage the inherent sparsity within attention mechanisms, both in conventional Softmax attention and ReLU attention (with $\mathsf{ReLU}^\alpha$ activation, $\alpha \in \mathbb{N}_+$), to significantly reduce the running time complexity. Our method employs a Half-Space Reporting (HSR) data structure to rapidly identify non-zero or "massively activated" entries in the attention matrix. We present theoretical analyses for two key scenarios: attention generation and full attention computation with long input context. Our approach achieves a running time of $O(mn^{4/5})$ significantly faster than the naive approach $O(mn)$ for attention generation, where $n$ is the context length, $m$ is the query length, and $d$ is the hidden dimension. We can also reduce the running time of full attention computation from $O(mn)$ to $O(mn^{1 - 1 / \lfloor d/2\rfloor} + mn^{4/5})$. Importantly, our method introduces no error for ReLU attention and only provably negligible error for Softmax attention, where the latter is supported by our empirical validation. This work represents a significant step towards enabling efficient long-context processing in LLMs, potentially broadening their applicability across various domains.

Scalable Weibull Graph Attention Autoencoder for Modeling Document Networks

Although existing variational graph autoencoders (VGAEs) have been widely used for modeling and generating graph-structured data, most of them are still not flexible enough to approximate the sparse and skewed latent node representations, especially those of document relational networks (DRNs) with discrete observations. To analyze a collection of interconnected documents, a typical branch of Bayesian models, specifically relational topic models (RTMs), has proven their efficacy in describing both link structures and document contents of DRNs, which motives us to incorporate RTMs with existing VGAEs to alleviate their potential issues when modeling the generation of DRNs. In this paper, moving beyond the sophisticated approximate assumptions of traditional RTMs, we develop a graph Poisson factor analysis (GPFA), which provides analytic conditional posteriors to improve the inference accuracy, and extend GPFA to a multi-stochastic-layer version named graph Poisson gamma belief network (GPGBN) to capture the hierarchical document relationships at multiple semantic levels. Then, taking GPGBN as the decoder, we combine it with various Weibull-based graph inference networks, resulting in two variants of Weibull graph auto-encoder (WGAE), equipped with model inference algorithms. Experimental results demonstrate that our models can extract high-quality hierarchical latent document representations and achieve promising performance on various graph analytic tasks.