Abstract:Massive Over-activation Yielded Uplifts(MOYU) is an inherent property of large language models, and dynamic activation(DA) based on the MOYU property is a clever yet under-explored strategy designed to accelerate inference in these models. Existing methods that utilize MOYU often face a significant 'Impossible Trinity': struggling to simultaneously maintain model performance, enhance inference speed, and extend applicability across various architectures. Due to the theoretical ambiguities surrounding MOYU, this paper elucidates the root cause of the MOYU property and outlines the mechanisms behind two primary limitations encountered by current DA methods: 1) history-related activation uncertainty, and 2) semantic-irrelevant activation inertia. Our analysis not only underscores the limitations of current dynamic activation strategies within large-scale LLaMA models but also proposes opportunities for refining the design of future sparsity schemes.
Abstract:Quantization is a proven effective method for compressing large language models. Although popular techniques like W8A8 and W4A16 effectively maintain model performance, they often fail to concurrently speed up the prefill and decoding stages of inference. W4A8 is a promising strategy to accelerate both of them while usually leads to a significant performance degradation. To address these issues, we present QQQ, a Quality Quattuor-bit Quantization method with 4-bit weights and 8-bit activations. QQQ employs adaptive smoothing and Hessian-based compensation, significantly enhancing the performance of quantized models without extensive training. Furthermore, we meticulously engineer W4A8 GEMM kernels to increase inference speed. Our specialized per-channel W4A8 GEMM and per-group W4A8 GEMM achieve impressive speed increases of 3.67$\times$ and 3.29 $\times$ over FP16 GEMM. Our extensive experiments show that QQQ achieves performance on par with existing state-of-the-art LLM quantization methods while significantly accelerating inference, achieving speed boosts up to 2.24 $\times$, 2.10$\times$, and 1.25$\times$ compared to FP16, W8A8, and W4A16, respectively.
Abstract:3D Gaussian Splatting-based techniques have recently advanced 3D scene reconstruction and novel view synthesis, achieving high-quality real-time rendering. However, these approaches are inherently limited by the underlying pinhole camera assumption in modeling the images and hence only work for All-in-Focus (AiF) sharp image inputs. This severely affects their applicability in real-world scenarios where images often exhibit defocus blur due to the limited depth-of-field (DOF) of imaging devices. Additionally, existing 3D Gaussian Splatting (3DGS) methods also do not support rendering of DOF effects. To address these challenges, we introduce DOF-GS that allows for rendering adjustable DOF effects, removing defocus blur as well as refocusing of 3D scenes, all from multi-view images degraded by defocus blur. To this end, we re-imagine the traditional Gaussian Splatting pipeline by employing a finite aperture camera model coupled with explicit, differentiable defocus rendering guided by the Circle-of-Confusion (CoC). The proposed framework provides for dynamic adjustment of DOF effects by changing the aperture and focal distance of the underlying camera model on-demand. It also enables rendering varying DOF effects of 3D scenes post-optimization, and generating AiF images from defocused training images. Furthermore, we devise a joint optimization strategy to further enhance details in the reconstructed scenes by jointly optimizing rendered defocused and AiF images. Our experimental results indicate that DOF-GS produces high-quality sharp all-in-focus renderings conditioned on inputs compromised by defocus blur, with the training process incurring only a modest increase in GPU memory consumption. We further demonstrate the applications of the proposed method for adjustable defocus rendering and refocusing of the 3D scene from input images degraded by defocus blur.
Abstract:Graph clustering is a fundamental problem in machine learning. Deep learning methods achieve the state-of-the-art results in recent years, but they still cannot work without predefined cluster numbers. Such limitation motivates us to pose a more challenging problem of graph clustering with unknown cluster number. We propose to address this problem from a fresh perspective of graph information theory (i.e., structural information). In the literature, structural information has not yet been introduced to deep clustering, and its classic definition falls short of discrete formulation and modeling node features. In this work, we first formulate a differentiable structural information (DSI) in the continuous realm, accompanied by several theoretical results. By minimizing DSI, we construct the optimal partitioning tree where densely connected nodes in the graph tend to have the same assignment, revealing the cluster structure. DSI is also theoretically presented as a new graph clustering objective, not requiring the predefined cluster number. Furthermore, we design a neural LSEnet in the Lorentz model of hyperbolic space, where we integrate node features to structural information via manifold-valued graph convolution. Extensive empirical results on real graphs show the superiority of our approach.
Abstract:In this work, we systematically investigate the efficacy of dynamic activation mechanisms within the LLaMA family of language models. Despite the potential of dynamic activation methods to reduce computation and increase speed in models using the ReLU activation function, our empirical findings have uncovered several inherent pitfalls in the current dynamic activation schemes. Through extensive experiments across various dynamic activation strategies, we demonstrate that LLaMA models usually underperform when compared to their ReLU counterparts, particularly in scenarios demanding high sparsity ratio. We attribute these deficiencies to a combination of factors: 1) the inherent complexity of dynamically predicting activation heads and neurons; 2) the inadequate sparsity resulting from activation functions; 3) the insufficient preservation of information resulting from KV cache skipping. Our analysis not only sheds light on the limitations of dynamic activation in the context of large-scale LLaMA models but also proposes roadmaps for enhancing the design of future sparsity schemes.
Abstract:Sentence Pattern Structure (SPS) parsing is a syntactic analysis method primarily employed in language teaching.Existing SPS parsers rely heavily on textbook corpora for training, lacking cross-domain capability.To overcome this constraint, this paper proposes an innovative approach leveraging large language models (LLMs) within a self-training framework. Partial syntactic rules from a source domain are combined with target domain sentences to dynamically generate training data, enhancing the adaptability of the parser to diverse domains.Experiments conducted on textbook and news domains demonstrate the effectiveness of the proposed method, outperforming rule-based baselines by 1.68 points on F1 metrics.
Abstract:Natural Language Processing (NLP) technologies have revolutionized the way we interact with information systems, with a significant focus on converting natural language queries into formal query languages such as SQL. However, less emphasis has been placed on the Corpus Query Language (CQL), a critical tool for linguistic research and detailed analysis within text corpora. The manual construction of CQL queries is a complex and time-intensive task that requires a great deal of expertise, which presents a notable challenge for both researchers and practitioners. This paper presents the first text-to-CQL task that aims to automate the translation of natural language into CQL. We present a comprehensive framework for this task, including a specifically curated large-scale dataset and methodologies leveraging large language models (LLMs) for effective text-to-CQL task. In addition, we established advanced evaluation metrics to assess the syntactic and semantic accuracy of the generated queries. We created innovative LLM-based conversion approaches and detailed experiments. The results demonstrate the efficacy of our methods and provide insights into the complexities of text-to-CQL task.
Abstract:This work introduces Weaver, our first family of large language models (LLMs) dedicated to content creation. Weaver is pre-trained on a carefully selected corpus that focuses on improving the writing capabilities of large language models. We then fine-tune Weaver for creative and professional writing purposes and align it to the preference of professional writers using a suit of novel methods for instruction data synthesis and LLM alignment, making it able to produce more human-like texts and follow more diverse instructions for content creation. The Weaver family consists of models of Weaver Mini (1.8B), Weaver Base (6B), Weaver Pro (14B), and Weaver Ultra (34B) sizes, suitable for different applications and can be dynamically dispatched by a routing agent according to query complexity to balance response quality and computation cost. Evaluation on a carefully curated benchmark for assessing the writing capabilities of LLMs shows Weaver models of all sizes outperform generalist LLMs several times larger than them. Notably, our most-capable Weaver Ultra model surpasses GPT-4, a state-of-the-art generalist LLM, on various writing scenarios, demonstrating the advantage of training specialized LLMs for writing purposes. Moreover, Weaver natively supports retrieval-augmented generation (RAG) and function calling (tool usage). We present various use cases of these abilities for improving AI-assisted writing systems, including integration of external knowledge bases, tools, or APIs, and providing personalized writing assistance. Furthermore, we discuss and summarize a guideline and best practices for pre-training and fine-tuning domain-specific LLMs.
Abstract:Large language models (LLMs) are incredibly powerful at comprehending and generating data in the form of text, but are brittle and error-prone. There has been an advent of toolkits and recipes centered around so-called prompt engineering-the process of asking an LLM to do something via a series of prompts. However, for LLM-powered data processing workflows, in particular, optimizing for quality, while keeping cost bounded, is a tedious, manual process. We put forth a vision for declarative prompt engineering. We view LLMs like crowd workers and leverage ideas from the declarative crowdsourcing literature-including leveraging multiple prompting strategies, ensuring internal consistency, and exploring hybrid-LLM-non-LLM approaches-to make prompt engineering a more principled process. Preliminary case studies on sorting, entity resolution, and imputation demonstrate the promise of our approach
Abstract:Transformer models have emerged as the leading approach for achieving state-of-the-art performance across various application domains, serving as the foundation for advanced large-scale deep learning (DL) models. However, efficiently training these models across multiple GPUs remains a complex challenge due to the abundance of parallelism options. Existing DL systems either require manual efforts to design distributed training plans or limit parallelism combinations to a constrained search space. In this paper, we present Galvatron-BMW, a novel system framework that integrates multiple prevalent parallelism dimensions and automatically identifies the most efficient hybrid parallelism strategy. To effectively navigate this vast search space, we employ a decision tree approach for decomposition and pruning based on intuitive insights. We further utilize a dynamic programming search algorithm to derive the optimal plan. Moreover, to improve resource utilization and enhance system efficiency, we propose a bi-objective optimization workflow that focuses on workload balance. Our evaluations on different Transformer models demonstrate the capabilities of Galvatron-BMW in automating distributed training under varying GPU memory constraints. Across all tested scenarios, Galvatron-BMW consistently achieves superior system throughput, surpassing previous approaches that rely on limited parallelism strategies.