The inference of ML models composed of diverse structures, types, and sizes boils down to the execution of different dataflows (i.e. different tiling, ordering, parallelism, and shapes). Using the optimal dataflow for every layer of workload can reduce latency by up to two orders of magnitude over a suboptimal dataflow. Unfortunately, reconfiguring hardware for different dataflows involves on-chip data layout reordering and datapath reconfigurations, leading to non-trivial overhead that hinders ML accelerators from exploiting different dataflows, resulting in suboptimal performance. To address this challenge, we propose FEATHER, an innovative accelerator that leverages a novel spatial array termed Nest and a novel multi-stage reduction network called BIRRD for performing flexible data reduction with layout reordering under the hood, enabling seamless switching between optimal dataflows with negligible latency and resources overhead. For systematically evaluating the performance interaction between dataflows and layouts, we enhance Timeloop, a state-of-the-art dataflow cost modeling and search framework, with layout assessment capabilities, and term it as Layoutloop. We model FEATHER into Layoutloop and also deploy FEATHER end-to-end on the edge ZCU104 FPGA. FEATHER delivers 1.27~2.89x inference latency speedup and 1.3~6.43x energy efficiency improvement compared to various SoTAs like NVDLA, SIGMA and Eyeriss under ResNet-50 and MobiletNet-V3 in Layoutloop. On practical FPGA devices, FEATHER achieves 2.65/3.91x higher throughput than Xilinx DPU/Gemmini. Remarkably, such performance and energy efficiency enhancements come at only 6% area over a fixed-dataflow Eyeriss-like accelerator. Our code is released at //github.com/maeri-project/FEATHER.
相關內容
Sparsely-activated Mixture-of-Expert (MoE) layers have found practical applications in enlarging the model size of large-scale foundation models, with only a sub-linear increase in computation demands. Despite the wide adoption of hybrid parallel paradigms like model parallelism, expert parallelism, and expert-sharding parallelism (i.e., MP+EP+ESP) to support MoE model training on GPU clusters, the training efficiency is hindered by communication costs introduced by these parallel paradigms. To address this limitation, we propose Parm, a system that accelerates MP+EP+ESP training by designing two dedicated schedules for placing communication tasks. The proposed schedules eliminate redundant computations and communications and enable overlaps between intra-node and inter-node communications, ultimately reducing the overall training time. As the two schedules are not mutually exclusive, we provide comprehensive theoretical analyses and derive an automatic and accurate solution to determine which schedule should be applied in different scenarios. Experimental results on an 8-GPU server and a 32-GPU cluster demonstrate that Parm outperforms the state-of-the-art MoE training system, DeepSpeed-MoE, achieving 1.13$\times$ to 5.77$\times$ speedup on 1296 manually configured MoE layers and approximately 3$\times$ improvement on two real-world MoE models based on BERT and GPT-2.
Recent work has explored the capability of large language models (LLMs) to identify and correct errors in LLM-generated responses. These refinement approaches frequently evaluate what sizes of models are able to do refinement for what problems, but less attention is paid to what effective feedback for refinement looks like. In this work, we propose looking at refinement with feedback as a composition of three distinct LLM competencies: (1) identification of bad generations; (2) fine-grained natural language feedback generation; (3) refining with fine-grained feedback. The first step can be implemented with a high-performing discriminative model and steps 2 and 3 can be implemented either via prompted or fine-tuned LLMs. A key property of this approach is that the step 2 critique model can give fine-grained feedback about errors, made possible by offloading the discrimination to a separate model in step 1. We show that models of different capabilities benefit from refining with this approach on the task of improving factual consistency of document grounded summaries. Overall, our proposed method consistently outperforms existing end-to-end refinement approaches and current trained models not fine-tuned for factuality critiquing.
Deployment of autoregressive large language models (LLMs) is costly, and as these models increase in size, the associated costs will become even more considerable. Consequently, different methods have been proposed to accelerate the token generation process and reduce costs. Speculative decoding (SD) is among the most promising approaches to speed up the LLM decoding process by verifying multiple tokens in parallel and using an auxiliary smaller draft model to generate the possible tokens. In SD, usually, one draft model is used to serve a specific target model; however, in practice, LLMs are diverse, and we might need to deal with many target models or more than one target model simultaneously. In this scenario, it is not clear which draft model should be used for which target model, and searching among different draft models or training customized draft models can further increase deployment costs. In this paper, we first introduce a novel multi-target scenario for the deployment of draft models for faster inference. Then, we present a novel, more efficient sorted speculative decoding mechanism that outperforms regular baselines in multi-target settings. We evaluated our method on Spec-Bench in different settings, including base models such as Vicuna 7B, 13B, and LLama Chat 70B. Our results suggest that our draft models perform better than baselines for multiple target models at the same time.
Neurosymbolic background knowledge and the expressivity required of its logic can break Machine Learning assumptions about data Independence and Identical Distribution. In this position paper we propose to analyze IID relaxation in a hierarchy of logics that fit different use case requirements. We discuss the benefits of exploiting known data dependencies and distribution constraints for Neurosymbolic use cases and argue that the expressivity required for this knowledge has implications for the design of underlying ML routines. This opens a new research agenda with general questions about Neurosymbolic background knowledge and the expressivity required of its logic.
The scalability of large language models (LLMs) in handling high-complexity models and large-scale datasets has led to tremendous successes in pivotal domains. While there is an urgent need to acquire more training data for LLMs, a concerning reality is the depletion of high-quality public datasets within a few years. In view of this, the federated learning (FL) LLM fine-tuning paradigm recently has been proposed to facilitate collaborative LLM fine-tuning on distributed private data, where multiple data owners collaboratively fine-tune a shared LLM without sharing raw data. However, the staggering model size of LLMs imposes heavy computing and communication burdens on clients, posing significant barriers to the democratization of the FL LLM fine-tuning paradigm. To address this issue, split learning (SL) has emerged as a promising solution by offloading the primary training workload to a server via model partitioning while exchanging activation/activation's gradients with smaller data sizes rather than the entire LLM. Unfortunately, research on the SL LLM fine-tuning paradigm is still in its nascent stage. To fill this gap, in this paper, we propose the first SL LLM fine-tuning framework, named SplitLoRA. SplitLoRA is built on the split federated learning (SFL) framework, amalgamating the advantages of parallel training from FL and model splitting from SL and thus greatly enhancing the training efficiency. It is worth noting that SplitLoRA is the inaugural open-source benchmark for SL LLM fine-tuning, providing a foundation for research efforts dedicated to advancing SL LLM fine-tuning. Extensive simulations validate that SplitLoRA achieves target accuracy in significantly less time than state-of-the-art LLM fine-tuning frameworks, demonstrating the superior training performance of SplitLoRA. The project page is available at //fduinc.github.io/splitlora/.
As opposed to evaluating computation and logic-based reasoning, current benchmarks for evaluating large language models (LLMs) in medicine are primarily focused on question-answering involving domain knowledge and descriptive reasoning. While such qualitative capabilities are vital to medical diagnosis, in real-world scenarios, doctors frequently use clinical calculators that follow quantitative equations and rule-based reasoning paradigms for evidence-based decision support. To this end, we propose MedCalc-Bench, a first-of-its-kind dataset focused on evaluating the medical calculation capability of LLMs. MedCalc-Bench contains an evaluation set of over 1000 manually reviewed instances from 55 different medical calculation tasks. Each instance in MedCalc-Bench consists of a patient note, a question requesting to compute a specific medical value, a ground truth answer, and a step-by-step explanation showing how the answer is obtained. While our evaluation results show the potential of LLMs in this area, none of them are effective enough for clinical settings. Common issues include extracting the incorrect entities, not using the correct equation or rules for a calculation task, or incorrectly performing the arithmetic for the computation. We hope our study highlights the quantitative knowledge and reasoning gaps in LLMs within medical settings, encouraging future improvements of LLMs for various clinical calculation tasks.
The recent success of large language models (LLMs) trained on static, pre-collected, general datasets has sparked numerous research directions and applications. One such direction addresses the non-trivial challenge of integrating pre-trained LLMs into dynamic data distributions, task structures, and user preferences. Pre-trained LLMs, when tailored for specific needs, often experience significant performance degradation in previous knowledge domains -- a phenomenon known as "catastrophic forgetting". While extensively studied in the continual learning (CL) community, it presents new manifestations in the realm of LLMs. In this survey, we provide a comprehensive overview of the current research progress on LLMs within the context of CL. This survey is structured into four main sections: we first describe an overview of continually learning LLMs, consisting of two directions of continuity: vertical continuity (or vertical continual learning), i.e., continual adaptation from general to specific capabilities, and horizontal continuity (or horizontal continual learning), i.e., continual adaptation across time and domains (Section 3). We then summarize three stages of learning LLMs in the context of modern CL: Continual Pre-Training (CPT), Domain-Adaptive Pre-training (DAP), and Continual Fine-Tuning (CFT) (Section 4). Then we provide an overview of evaluation protocols for continual learning with LLMs, along with the current available data sources (Section 5). Finally, we discuss intriguing questions pertaining to continual learning for LLMs (Section 6). The full list of papers examined in this survey is available at //github.com/Wang-ML-Lab/llm-continual-learning-survey.
Reduced order models (ROMs) have achieved a lot of success in reducing the computational cost of traditional numerical methods across many disciplines. For convection-dominated (e.g., turbulent) flows, however, standard ROMs generally yield inaccurate results, usually affected by spurious oscillations. Thus, ROMs are usually equipped with numerical stabilization or closure models to account for the effect of the discarded modes. The literature on ROM closures and stabilizations is large and growing fast. In this paper, we focus on one particular type of ROM closures and stabilizations that are inspired by Large Eddy Simulation (LES). These ROMs, which we call LES-ROMs, are extremely easy to implement, very efficient, and accurate. Carefully tuned LES-ROMs can accurately capture the average physical quantities of interest in challenging convection-dominated flows in many applications. LES-ROM are constructed by leveraging spatial filtering, i.e., the same principle used to build classical LES models. This ensures a modeling consistency between LES-ROMs and the approaches that generated the data used to train them. It also ``bridges'' two distinct research fields (LES and ROMs), disconnected until now. This paper is a review of LES-ROMs. It starts with a description of a versatile LES strategy called evolve-filter-relax (EFR) that has been successfully used as a full order method. We then show how the EFR strategy, and spatial filtering in general, can be leveraged to construct LES-ROMs. Several applications of LES-ROMs are presented. Finally, we draw conclusions and outline several research directions and open questions in the LES-ROM development. While we do not claim this review to be comprehensive, we certainly hope it serves as a brief and friendly introduction to this exciting research area, which has a lot of potential in practical numerical simulation of convection-dominated flows.
External influences such as traffic and environmental factors induce vibrations in structures, leading to material degradation over time. These vibrations result in cracks due to the material's lack of plasticity compromising structural integrity. Detecting such damage requires the installation of vibration sensors to capture the internal dynamics. However, distinguishing relevant eigenmodes from external noise necessitates the use of Deep Learning models. The detection of changes in eigenmodes can be used to anticipate these shifts in material properties and to discern between normal and anomalous structural behavior. Eigenmodes, representing characteristic vibration patterns, provide insights into structural dynamics and deviations from expected states. Thus, we propose ModeConv to automatically capture and analyze changes in eigenmodes, facilitating effective anomaly detection in structures and material properties. In the conducted experiments, ModeConv demonstrates computational efficiency improvements, resulting in reduced runtime for model calculations. The novel ModeConv neural network layer is tailored for temporal graph neural networks, in which every node represents one sensor. ModeConv employs a singular value decomposition based convolutional filter design for complex numbers and leverages modal transformation in lieu of Fourier or Laplace transformations in spectral graph convolutions. We include a mathematical complexity analysis illustrating the runtime reduction.
The emergence of large language models (LLMs) has substantially influenced natural language processing, demonstrating exceptional results across various tasks. In this study, we employ ``Introspective Tips" to facilitate LLMs in self-optimizing their decision-making. By introspectively examining trajectories, LLM refines its policy by generating succinct and valuable tips. Our method enhances the agent's performance in both few-shot and zero-shot learning situations by considering three essential scenarios: learning from the agent's past experiences, integrating expert demonstrations, and generalizing across diverse games. Importantly, we accomplish these improvements without fine-tuning the LLM parameters; rather, we adjust the prompt to generalize insights from the three aforementioned situations. Our framework not only supports but also emphasizes the advantage of employing LLM in in-contxt decision-making. Experiments involving over 100 games in TextWorld illustrate the superior performance of our approach.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191