The k-means algorithm can simplify large-scale spatial vectors, such as 2D geo-locations and 3D point clouds, to support fast analytics and learning. However, when processing large-scale datasets, existing k-means algorithms have been developed to achieve high performance with significant computational resources, such as memory and CPU usage time. These algorithms, though effective, are not well-suited for resource-constrained devices. In this paper, we propose a fast, memory-efficient, and cost-predictable k-means called Dask-means. We first accelerate k-means by designing a memory-efficient accelerator, which utilizes an optimized nearest neighbor search over a memory-tunable index to assign spatial vectors to clusters in batches. We then design a lightweight cost estimator to predict the memory cost and runtime of the k-means task, allowing it to request appropriate memory from devices or adjust the accelerator's required space to meet memory constraints, and ensure sufficient CPU time for running k-means. Experiments show that when simplifying datasets with scale such as $10^6$, Dask-means uses less than $30$MB of memory, achieves over $168$ times speedup compared to the widely-used Lloyd's algorithm. We also validate Dask-means on mobile devices, where it demonstrates significant speedup and low memory cost compared to other state-of-the-art (SOTA) k-means algorithms. Our cost estimator estimates the memory cost with a difference of less than $3\%$ from the actual ones and predicts runtime with an MSE up to $33.3\%$ lower than SOTA methods.
Constructing datasets representative of the target domain is essential for training effective machine learning models. Active learning (AL) is a promising method that iteratively extends training data to enhance model performance while minimizing data acquisition costs. However, current AL workflows often require human intervention and lack parallelism, leading to inefficiencies and underutilization of modern computational resources. In this work, we introduce PAL, an automated, modular, and parallel active learning library that integrates AL tasks and manages their execution and communication on shared- and distributed-memory systems using the Message Passing Interface (MPI). PAL provides users with the flexibility to design and customize all components of their active learning scenarios, including machine learning models with uncertainty estimation, oracles for ground truth labeling, and strategies for exploring the target space. We demonstrate that PAL significantly reduces computational overhead and improves scalability, achieving substantial speed-ups through asynchronous parallelization on CPU and GPU hardware. Applications of PAL to several real-world scenarios - including ground-state reactions in biomolecular systems, excited-state dynamics of molecules, simulations of inorganic clusters, and thermo-fluid dynamics - illustrate its effectiveness in accelerating the development of machine learning models. Our results show that PAL enables efficient utilization of high-performance computing resources in active learning workflows, fostering advancements in scientific research and engineering applications.
We introduce an algorithm for efficiently representing convolution with zero-padding and stride as a sparse transformation matrix, applied to a vectorized input through sparse matrix-vector multiplication (SpMV). We provide a theoretical contribution with an explicit expression for the number of non-zero multiplications in convolutions with stride and padding, offering insight into the potential for leveraging sparsity in convolution operations. A proof-of-concept implementation is presented in Python, demonstrating the performance of our method on both CPU and GPU architectures. This work contributes to the broader exploration of sparse matrix techniques in convolutional algorithms, with a particular focus on leveraging matrix multiplications for parallelization. Our findings lay the groundwork for future advancements in exploiting sparsity to improve the efficiency of convolution operations in fields such as machine learning and signal processing.
Transformer neural networks (TNN) excel in natural language processing (NLP), machine translation, and computer vision (CV) without relying on recurrent or convolutional layers. However, they have high computational and memory demands, particularly on resource-constrained devices like FPGAs. Moreover, transformer models vary in processing time across applications, requiring custom models with specific parameters. Designing custom accelerators for each model is complex and time-intensive. Some custom accelerators exist with no runtime adaptability, and they often rely on sparse matrices to reduce latency. However, hardware designs become more challenging due to the need for application-specific sparsity patterns. This paper introduces ADAPTOR, a runtime-adaptive accelerator for dense matrix computations in transformer encoders and decoders on FPGAs. ADAPTOR enhances the utilization of processing elements and on-chip memory, enhancing parallelism and reducing latency. It incorporates efficient matrix tiling to distribute resources across FPGA platforms and is fully quantized for computational efficiency and portability. Evaluations on Xilinx Alveo U55C data center cards and embedded platforms like VC707 and ZCU102 show that our design is 1.2$\times$ and 2.87$\times$ more power efficient than the NVIDIA K80 GPU and the i7-8700K CPU respectively. Additionally, it achieves a speedup of 1.7 to 2.25$\times$ compared to some state-of-the-art FPGA-based accelerators.
Inference for Large Language Models (LLMs) is computationally demanding. To reduce the cost of auto-regressive decoding, Key-Value (KV) cache is used to store intermediate activations, which significantly lowers the computational overhead for token generation. However, the memory required for the KV cache grows rapidly, often exceeding the capacity of GPU memory. A cost-effective alternative is to offload KV cache to CPU memory, which alleviates GPU memory pressure, but shifts the bottleneck to the limited bandwidth of the PCIe connection between the CPU and GPU. Existing methods attempt to address these issues by overlapping GPU computation with I/O or employing CPU-GPU heterogeneous execution, but they are hindered by excessive data movement and dependence on CPU capabilities. Fully overlapping PCIe communication latency gets challenging as the size of the KV cache grows and/or the GPU compute capabilities increase. In this paper, we introduce KVPR, an efficient I/O-aware LLM inference method where the CPU first transfers a partial set of activations, from which the GPU can start recomputing the KV cache values. While the GPU recomputes the partial KV cache, the remaining portion of the KV cache is transferred concurrently from the CPU. This approach overlaps GPU recomputation with KV cache transfer to minimize idle GPU time and maximize inference performance. KVPR is fully automated by integrating a profiler module that utilizes input characteristics and system hardware information, a scheduler module to optimize the distribution of computation and communication workloads, and a runtime module to efficiently execute the derived execution plan. Experimental results show that KVPR achieves up to 35.8% lower latency and 46.2% higher throughput during decoding compared to state-of-the-art approaches. The code is available at https://github.com/chaoyij/KVPR.
Arbor is a software library designed for efficient simulation of large-scale networks of biological neurons with detailed morphological structures. It combines customizable neuronal and synaptic mechanisms with high-performance computing, supporting multi-core CPU and GPU systems. In humans and other animals, synaptic plasticity processes play a vital role in cognitive functions, including learning and memory. Recent studies have shown that intracellular molecular processes in dendrites significantly influence single-neuron dynamics. However, for understanding how the complex interplay between dendrites and synaptic processes influences network dynamics, computational modeling is required. To enable the modeling of large-scale networks of morphologically detailed neurons with diverse plasticity processes, we have extended the Arbor library to the Plastic Arbor framework, supporting simulations of a large variety of spike-driven plasticity paradigms. To showcase the features of the new framework, we present examples of computational models, beginning with single-synapse dynamics, progressing to multi-synapse rules, and finally scaling up to large recurrent networks. While cross-validating our implementations by comparison with other simulators, we show that Arbor allows simulating plastic networks of multi-compartment neurons at nearly no additional cost in runtime compared to point-neuron simulations. Using the new framework, we have already been able to investigate the impact of dendritic structures on network dynamics across a timescale of several hours, showing a relation between the length of dendritic trees and the ability of the network to efficiently store information. By our extension of Arbor, we aim to provide a valuable tool that will support future studies on the impact of synaptic plasticity, especially, in conjunction with neuronal morphology, in large networks.
Deep learning models often require specially designed architectures to process data of different dimensions, such as 1D time series, 2D images, and 3D volumetric data. Existing bidirectional models mainly focus on sequential data, making it difficult to scale effectively to higher dimensions. To address this issue, we propose a novel multi-dimensional bidirectional neural network architecture, named Nd-BiMamba2, which efficiently handles 1D, 2D, and 3D data. Nd-BiMamba2 is based on the Mamba2 module and introduces innovative bidirectional processing mechanisms and adaptive padding strategies to capture bidirectional information in multi-dimensional data while maintaining computational efficiency. Unlike existing methods that require designing specific architectures for different dimensional data, Nd-BiMamba2 adopts a unified architecture with a modular design, simplifying development and maintenance costs. To verify the portability and flexibility of Nd-BiMamba2, we successfully exported it to ONNX and TorchScript and tested it on different hardware platforms (e.g., CPU, GPU, and mobile devices). Experimental results show that Nd-BiMamba2 runs efficiently on multiple platforms, demonstrating its potential in practical applications. The code is open-source: https://github.com/Human9000/nd-Mamba2-torch
We present a diffusion-based approach to quadrupedal locomotion that simultaneously addresses the limitations of learning and interpolating between multiple skills and of (modes) offline adapting to new locomotion behaviours after training. This is the first framework to apply classifier-free guided diffusion to quadruped locomotion and demonstrate its efficacy by extracting goal-conditioned behaviour from an originally unlabelled dataset. We show that these capabilities are compatible with a multi-skill policy and can be applied with little modification and minimal compute overhead, i.e., running entirely on the robots onboard CPU. We verify the validity of our approach with hardware experiments on the ANYmal quadruped platform.
In this paper, we derive a new Kalman filter with probabilistic data association between measurements and states. We formulate a variational inference problem to approximate the posterior density of the state conditioned on the measurement data. We view the unknown data association as a latent variable and apply Expectation Maximization (EM) to obtain a filter with update step in the same form as the Kalman filter but with expanded measurement vector of all potential associations. We show that the association probabilities can be computed as permanents of matrices with measurement likelihood entries. We also propose an ambiguity check that associates only a subset of ambiguous measurements and states probabilistically, thus reducing the association time and preventing low-probability measurements from harming the estimation accuracy. Experiments in simulation show that our filter achieves lower tracking errors than the well-established joint probabilistic data association filter (JPDAF), while running at comparable rate. We also demonstrate the effectiveness of our filter in multi-object tracking (MOT) on multiple real-world datasets, including MOT17, MOT20, and DanceTrack. We achieve better higher order tracking accuracy (HOTA) than previous Kalman-filter methods and remain real-time. Associating only bounding boxes without deep features or velocities, our method ranks top-10 on both MOT17 and MOT20 in terms of HOTA. Given offline detections, our algorithm tracks at 250+ fps on a single laptop CPU. Code is available at https://github.com/hwcao17/pkf.
Recent sustainability drives place energy-consumption metrics in centre-stage for the design of future radio access networks (RAN). At the same time, optimising the trade-off between performance and system energy usage by machine-learning (ML) is an approach that requires large amounts of granular RAN data to train models, and to adapt in near realtime. In this paper, we present extensions to the system-level discrete-event AIMM (AI-enabled Massive MIMO) Simulator, generating realistic figures for throughput and energy efficiency (EE) towards digital twin network modelling. We further investigate the trade-off between maximising either EE or spectrum efficiency (SE). To this end, we have run extensive simulations of a typical macrocell network deployment under various transmit power-reduction scenarios with a range of difference of 43 dBm. Our results demonstrate that the EE and SE objectives often require different power settings in different scenarios. Importantly, low mean user CPU execution times of 2.17 $\pm$ 0.05 seconds (2~s.d.) demonstrate that the AIMM Simulator is a powerful tool for quick prototyping of scalable system models which can interface with ML frameworks, and thus support future research in energy-efficient next generation networks.
Bayesian Additive Regression Trees (BART) is a nonparametric Bayesian regression technique based on an ensemble of decision trees. It is part of the toolbox of many statisticians. The overall statistical quality of the regression is typically higher than other generic alternatives, and it requires less manual tuning, making it a good default choice. However, it is a niche method compared to its natural competitor XGBoost, due to the longer running time, making sample sizes above 10,000-100,000 a nuisance. I present a GPU-enabled implementation of BART, faster by up to 200x relative to a single CPU core, making BART competitive in running time with XGBoost. This implementation is available in the Python package bartz.
Transformers and large language models~(LLMs) have seen rapid adoption in all domains. Their sizes have exploded to hundreds of billions of parameters and keep increasing. Under these circumstances, the training of transformers is very expensive and often hits a ``memory wall'', i.e., even when using 3D parallelism (pipeline, tensor, data) and aggregating the memory of many GPUs, it is still not enough to hold the necessary data structures (model parameters, optimizer state, gradients, activations) in GPU memory. To compensate, state-of-the-art approaches offload the optimizer state, at least partially, to the host memory and perform hybrid CPU-GPU computations. However, the management of the combined host-GPU memory is often suboptimal and results in poor overlapping between data movements and computations. This leads to missed opportunities to simultaneously leverage the interconnect bandwidth and computational capabilities of CPUs and GPUs. In this paper, we leverage a key observation that the interleaving of the forward, backward and update phases generate fluctuations in the GPU memory utilization, which can be exploited to dynamically move a part of the optimizer state between the host and the GPU memory at each iteration. To this end, we design and implement \proj, a novel technique to split the LLM into subgroups, whose update phase is scheduled on either the CPU or the GPU based on our proposed performance model that addresses the trade-off between data movement cost, acceleration on the GPUs vs the CPUs, and competition for shared resources. We integrate our approach with DeepSpeed and demonstrate 2.5$\times$ faster iterations over state-of-the-art approaches using extensive experiments.
With the widespread deployment of long-context large language models (LLMs), there has been a growing demand for efficient support of high-throughput inference. However, as the key-value (KV) cache expands with the sequence length, the increasing memory footprint and the need to access it for each token generation both result in low throughput when serving long-context LLMs. While various dynamic sparse attention methods have been proposed to speed up inference while maintaining generation quality, they either fail to sufficiently reduce GPU memory consumption or introduce significant decoding latency by offloading the KV cache to the CPU. We present ShadowKV, a high-throughput long-context LLM inference system that stores the low-rank key cache and offloads the value cache to reduce the memory footprint for larger batch sizes and longer sequences. To minimize decoding latency, ShadowKV employs an accurate KV selection strategy that reconstructs minimal sparse KV pairs on-the-fly. By evaluating ShadowKV on a broad range of benchmarks, including RULER, LongBench, and Needle In A Haystack, and models like Llama-3.1-8B, Llama-3-8B-1M, GLM-4-9B-1M, Yi-9B-200K, Phi-3-Mini-128K, and Qwen2-7B-128K, we demonstrate that it can support up to 6$\times$ larger batch sizes and boost throughput by up to 3.04$\times$ on an A100 GPU without sacrificing accuracy, even surpassing the performance achievable with infinite batch size under the assumption of infinite GPU memory. The code is available at https://github.com/bytedance/ShadowKV.
Raw nanopore signal analysis is a common approach in genomics to provide fast and resource-efficient analysis without translating the signals to bases (i.e., without basecalling). However, existing solutions cannot interpret raw signals directly if a reference genome is unknown due to a lack of accurate mechanisms to handle increased noise in pairwise raw signal comparison. Our goal is to enable the direct analysis of raw signals without a reference genome. To this end, we propose Rawsamble, the first mechanism that can 1) identify regions of similarity between all raw signal pairs, known as all-vs-all overlapping, using a hash-based search mechanism and 2) use these to construct genomes from scratch, called de novo assembly. Our extensive evaluations across multiple genomes of varying sizes show that Rawsamble provides a significant speedup (on average by 16.36x and up to 41.59x) and reduces peak memory usage (on average by 11.73x and up to by 41.99x) compared to a conventional genome assembly pipeline using the state-of-the-art tools for basecalling (Dorado's fastest mode) and overlapping (minimap2) on a CPU. We find that 36.57% of overlapping pairs generated by Rawsamble are identical to those generated by minimap2. Using the overlaps from Rawsamble, we construct the first de novo assemblies directly from raw signals without basecalling. We show that we can construct contiguous assembly segments (unitigs) up to 2.7 million bases in length (half the genome length of E. coli). We identify previously unexplored directions that can be enabled by finding overlaps and constructing de novo assemblies. Rawsamble is available at https://github.com/CMU-SAFARI/RawHash. We also provide the scripts to fully reproduce our results on our GitHub page.
Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by up to $5\times$ across various GPU hardware and achieve 54ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens. The code is available at https://github.com/Infini-AI-Lab/MagicPIG.
Large Language Models (LLMs) have emerged as powerful tools in mathematical theorem proving, particularly when utilizing formal languages such as LEAN. The major learning paradigm is expert iteration, which necessitates a pre-defined dataset comprising numerous mathematical problems. In this process, LLMs attempt to prove problems within the dataset and iteratively refine their capabilities through self-training on the proofs they discover. We propose to use large scale LEAN problem datasets Lean-workbook for expert iteration with more than 20,000 CPU days. During expert iteration, we found log-linear trends between solved problem amount with proof length and CPU usage. We train a critic model to select relatively easy problems for policy models to make trials and guide the model to search for deeper proofs. InternLM2.5-StepProver achieves open-source state-of-the-art on MiniF2F, Lean-Workbook-Plus, ProofNet, and Putnam benchmarks. Specifically, it achieves a pass of 65.9% on the MiniF2F-test and proves (or disproves) 17.0% of problems in Lean-Workbook-Plus which shows a significant improvement compared to only 9.5% of problems proved when Lean-Workbook-Plus was released. We open-source our models and searched proofs at https://github.com/InternLM/InternLM-Math and https://huggingface.co/datasets/internlm/Lean-Workbook.
Current and future large scale structure surveys aim to constrain the neutrino mass and the equation of state of dark energy. We aim to construct accurate and interpretable symbolic approximations to the linear and nonlinear matter power spectra as a function of cosmological parameters in extended $\Lambda$CDM models which contain massive neutrinos and non-constant equations of state for dark energy. This constitutes an extension of the syren-halofit emulators to incorporate these two effects, which we call syren-new (SYmbolic-Regression-ENhanced power spectrum emulator with NEutrinos and $W_0-w_a$). We also obtain a simple approximation to the derived parameter $\sigma_8$ as a function of the cosmological parameters for these models. Our results for the linear power spectrum are designed to emulate CLASS, whereas for the nonlinear case we aim to match the results of EuclidEmulator2. We compare our results to existing emulators and $N$-body simulations. Our analytic emulators for $\sigma_8$, the linear and nonlinear power spectra achieve root mean squared errors of 0.1%, 0.3% and 1.3%, respectively, across a wide range of cosmological parameters, redshifts and wavenumbers. We verify that emulator-related discrepancies are subdominant compared to observational errors and other modelling uncertainties when computing shear power spectra for LSST-like surveys. Our expressions have similar accuracy to existing (numerical) emulators, but are at least an order of magnitude faster, both on a CPU and GPU. Our work greatly improves the accuracy, speed and range of applicability of current symbolic approximations to the linear and nonlinear matter power spectra. We provide publicly available code for all symbolic approximations found.
Two-stage object detectors exhibit high accuracy and precise localization, especially for identifying small objects that are favorable for various edge applications. However, the high computation costs associated with two-stage detection methods cause more severe thermal issues on edge devices, incurring dynamic runtime frequency change and thus large inference latency variations. Furthermore, the dynamic number of proposals in different frames leads to various computations over time, resulting in further latency variations. The significant latency variations of detectors on edge devices can harm user experience and waste hardware resources. To avoid thermal throttling and provide stable inference speed, we propose Lotus, a novel framework that is tailored for two-stage detectors to dynamically scale CPU and GPU frequencies jointly in an online manner based on deep reinforcement learning (DRL). To demonstrate the effectiveness of Lotus, we implement it on NVIDIA Jetson Orin Nano and Mi 11 Lite mobile platforms. The results indicate that Lotus can consistently and significantly reduce latency variation, achieve faster inference, and maintain lower CPU and GPU temperatures under various settings.
Inference with transformer-based language models begins with a prompt processing step. In this step, the model generates the first output token and stores the KV cache needed for future generation steps. This prompt processing step can be computationally expensive, taking 10s of seconds or more for billion-parameter models on edge devices when prompt lengths or batch sizes rise. This degrades user experience by introducing significant latency into the model's outputs. To reduce the time spent producing the first output (known as the ``time to first token'', or TTFT) of a pretrained model, we introduce a novel method called KV Prediction. In our method, a small auxiliary model is used to process the prompt and produce an approximation of the KV cache used by a base model. This approximated KV cache is then used with the base model for autoregressive generation without the need to query the auxiliary model again. We demonstrate that our method produces a pareto-optimal efficiency-accuracy trade-off when compared to baselines. On TriviaQA, we demonstrate relative accuracy improvements in the range of $15\%-50\%$ across a range of TTFT FLOPs budgets. We also demonstrate accuracy improvements of up to $30\%$ on HumanEval python code completion at fixed TTFT FLOPs budgets. Additionally, we benchmark models on an Apple M2 Pro CPU and demonstrate that our improvement in FLOPs translates to a TTFT speedup on hardware. We release our code at https://github.com/apple/corenet/tree/main/projects/kv-prediction .
Explainable molecular property prediction is essential for various scientific fields, such as drug discovery and material science. Despite delivering intrinsic explainability, linear models struggle with capturing complex, non-linear patterns. Large language models (LLMs), on the other hand, yield accurate predictions through powerful inference capabilities yet fail to provide chemically meaningful explanations for their predictions. This work proposes a novel framework, called MoleX, which leverages LLM knowledge to build a simple yet powerful linear model for accurate molecular property prediction with faithful explanations. The core of MoleX is to model complicated molecular structure-property relationships using a simple linear model, augmented by LLM knowledge and a crafted calibration strategy. Specifically, to extract the maximum amount of task-relevant knowledge from LLM embeddings, we employ information bottleneck-inspired fine-tuning and sparsity-inducing dimensionality reduction. These informative embeddings are then used to fit a linear model for explainable inference. Moreover, we introduce residual calibration to address prediction errors stemming from linear models' insufficient expressiveness of complex LLM embeddings, thus recovering the LLM's predictive power and boosting overall accuracy. Theoretically, we provide a mathematical foundation to justify MoleX's explainability. Extensive experiments demonstrate that MoleX outperforms existing methods in molecular property prediction, establishing a new milestone in predictive performance, explainability, and efficiency. In particular, MoleX enables CPU inference and accelerates large-scale dataset processing, achieving comparable performance 300x faster with 100,000 fewer parameters than LLMs. Additionally, the calibration improves model performance by up to 12.7% without compromising explainability.