Search

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

We introduce Speech ReaLLM, a new ASR architecture that marries "decoder-only" ASR with the RNN-T to make multimodal LLM architectures capable of real-time streaming. This is the first "decoder-only" ASR architecture designed to handle continuous audio without explicit end-pointing. Speech ReaLLM is a special case of the more general ReaLLM ("real-time LLM") approach, also introduced here for the first time. The idea is inspired by RNN-T: Instead of generating a response only at the end of a user prompt, generate after every input token received in real time (it is often empty). On Librispeech "test", an 80M Speech ReaLLM achieves WERs of 3.0% and 7.4% in real time (without an external LM or auxiliary loss). This is only slightly above a 3x larger Attention-Encoder-Decoder baseline. We also show that this way, an LLM architecture can learn to represent and reproduce the flow of time; and that a pre-trained 7B LLM can be fine-tuned to do reasonably well on this task.

In this work, we extend the instruction-tuned Llama-2 model with end-to-end general-purpose speech processing and reasoning abilities while maintaining the wide range of original LLM capabilities, without using any carefully curated paired data. The resulting end-to-end model, named AudioChatLlama, can utilize audio prompts as a replacement for text and sustain a conversation. Such a model also has extended cross-modal capabilities such as being able to perform spoken question answering (QA), speech translation, and audio summarization amongst many other closed and open-domain tasks. This is unlike prior approaches in speech, in which LLMs are extended to handle audio for a limited number of pre-designated tasks. On both synthesized and recorded speech QA test sets, evaluations show that our end-to-end approach is on par with or outperforms cascaded systems (speech recognizer + LLM) in terms of modeling the response to a prompt. Furthermore, unlike cascades, our approach can interchange text and audio modalities and intrinsically utilize prior context in a conversation to provide better results.

Neural network pruning offers an effective method for compressing a multilingual automatic speech recognition (ASR) model with minimal performance loss. However, it entails several rounds of pruning and re-training needed to be run for each language. In this work, we propose the use of an adaptive masking approach in two scenarios for pruning a multilingual ASR model efficiently, each resulting in sparse monolingual models or a sparse multilingual model (named as Dynamic ASR Pathways). Our approach dynamically adapts the sub-network, avoiding premature decisions about a fixed sub-network structure. We show that our approach outperforms existing pruning methods when targeting sparse monolingual models. Further, we illustrate that Dynamic ASR Pathways jointly discovers and trains better sub-networks (pathways) of a single multilingual model by adapting from different sub-network initializations, thereby reducing the need for language-specific pruning.

Automatic Speech Recognition (ASR) models need to be optimized for specific hardware before they can be deployed on devices. This can be done by tuning the model's hyperparameters or exploring variations in its architecture. Re-training and re-validating models after making these changes can be a resource-intensive task. This paper presents TODM (Train Once Deploy Many), a new approach to efficiently train many sizes of hardware-friendly on-device ASR models with comparable GPU-hours to that of a single training job. TODM leverages insights from prior work on Supernet, where Recurrent Neural Network Transducer (RNN-T) models share weights within a Supernet. It reduces layer sizes and widths of the Supernet to obtain subnetworks, making them smaller models suitable for all hardware types. We introduce a novel combination of three techniques to improve the outcomes of the TODM Supernet: adaptive dropouts, an in-place Alpha-divergence knowledge distillation, and the use of ScaledAdam optimizer. We validate our approach by comparing Supernet-trained versus individually tuned Multi-Head State Space Model (MH-SSM) RNN-T using LibriSpeech. Results demonstrate that our TODM Supernet either matches or surpasses the performance of manually tuned models by up to a relative of 3% better in word error rate (WER), while efficiently keeping the cost of training many models at a small constant., Comment: Meta AI; Submitted to ICASSP 2024

Large language models have proven themselves highly flexible, able to solve a wide range of generative tasks, such as abstractive summarization and open-ended question answering. In this paper we extend the capabilities of LLMs by directly attaching a small audio encoder allowing it to perform speech recognition. By directly prepending a sequence of audial embeddings to the text token embeddings, the LLM can be converted to an automatic speech recognition (ASR) system, and be used in the exact same manner as its textual counterpart. Experiments on Multilingual LibriSpeech (MLS) show that incorporating a conformer encoder into the open sourced LLaMA-7B allows it to outperform monolingual baselines by 18% and perform multilingual speech recognition despite LLaMA being trained overwhelmingly on English text. Furthermore, we perform ablation studies to investigate whether the LLM can be completely frozen during training to maintain its original capabilities, scaling up the audio encoder, and increasing the audio encoder striding to generate fewer embeddings. The results from these studies show that multilingual ASR is possible even when the LLM is frozen or when strides of almost 1 second are used in the audio encoder opening up the possibility for LLMs to operate on long-form audio.

State space models (SSMs) have recently shown promising results on small-scale sequence and language modelling tasks, rivalling and outperforming many attention-based approaches. In this paper, we propose a multi-head state space (MH-SSM) architecture equipped with special gating mechanisms, where parallel heads are taught to learn local and global temporal dynamics on sequence data. As a drop-in replacement for multi-head attention in transformer encoders, this new model significantly outperforms the transformer transducer on the LibriSpeech speech recognition corpus. Furthermore, we augment the transformer block with MH-SSMs layers, referred to as the Stateformer, achieving state-of-the-art performance on the LibriSpeech task, with word error rates of 1.76\%/4.37\% on the development and 1.91\%/4.36\% on the test sets without using an external language model., Comment: Interspeech 2023

Neural transducers have achieved human level performance on standard speech recognition benchmarks. However, their performance significantly degrades in the presence of cross-talk, especially when the primary speaker has a low signal-to-noise ratio. Anchored speech recognition refers to a class of methods that use information from an anchor segment (e.g., wake-words) to recognize device-directed speech while ignoring interfering background speech. In this paper, we investigate anchored speech recognition to make neural transducers robust to background speech. We extract context information from the anchor segment with a tiny auxiliary network, and use encoder biasing and joiner gating to guide the transducer towards the target speech. Moreover, to improve the robustness of context embedding extraction, we propose auxiliary training objectives to disentangle lexical content from speaking style. We evaluate our methods on synthetic LibriSpeech-based mixtures comprising several SNR and overlap conditions; they improve relative word error rates by 19.6% over a strong baseline, when averaged over all conditions., Comment: To appear at IEEE ICASSP 2023

Graph neural networks (GNNs) have shown superiority in many prediction tasks over graphs due to their impressive capability of capturing nonlinear relations in graph-structured data. However, for node classification tasks, often, only marginal improvement of GNNs over their linear counterparts has been observed. Previous works provide very few understandings of this phenomenon. In this work, we resort to Bayesian learning to deeply investigate the functions of non-linearity in GNNs for node classification tasks. Given a graph generated from the statistical model CSBM, we observe that the max-a-posterior estimation of a node label given its own and neighbors' attributes consists of two types of non-linearity, a possibly non-linear transformation of node attributes and a ReLU-activated feature aggregation from neighbors. The latter surprisingly matches the type of non-linearity used in many GNN models. By further imposing Gaussian assumption on node attributes, we prove that the superiority of those ReLU activations is only significant when the node attributes are far more informative than the graph structure, which nicely matches many previous empirical observations. A similar argument can be achieved when there is a distribution shift of node attributes between the training and testing datasets. Finally, we verify our theory on both synthetic and real-world networks.

Cross-device federated learning (FL) protects user privacy by collaboratively training a model on user devices, therefore eliminating the need for collecting, storing, and manually labeling user data. While important topics such as the FL training algorithm, non-IID-ness, and Differential Privacy have been well studied in the literature, this paper focuses on two challenges of practical importance for improving on-device ASR: the lack of ground-truth transcriptions and the scarcity of compute resource and network bandwidth on edge devices. First, we propose a FL system for on-device ASR domain adaptation with full self-supervision, which uses self-labeling together with data augmentation and filtering techniques. The system can improve a strong Emformer-Transducer based ASR model pretrained on out-of-domain data, using in-domain audio without any ground-truth transcriptions. Second, to reduce the training cost, we propose a self-restricted RNN Transducer (SR-RNN-T) loss, a variant of alignment-restricted RNN-T that uses Viterbi alignments from self-supervision. To further reduce the compute and network cost, we systematically explore adapting only a subset of weights in the Emformer-Transducer. Our best training recipe achieves a $12.9\%$ relative WER reduction over the strong out-of-domain baseline, which equals $70\%$ of the reduction achievable with full human supervision and centralized training.

With the wide-spread availability of complex relational data, semi-supervised node classification in graphs has become a central machine learning problem. Graph neural networks are a recent class of easy-to-train and accurate methods for this problem that map the features in the neighborhood of a node to its label, but they ignore label correlation during inference and their predictions are difficult to interpret. On the other hand, collective classification is a traditional approach based on interpretable graphical models that explicitly model label correlations. Here, we introduce a model that combines the advantages of these two approaches, where we compute the marginal probabilities in a conditional random field, similar to collective classification, and the potentials in the random field are learned through end-to-end training, akin to graph neural networks. In our model, potentials on each node only depend on that node's features, and edge potentials are learned via a coupling matrix. This structure enables simple training with interpretable parameters, scales to large networks, naturally incorporates training labels at inference, and is often more accurate than related approaches. Our approach can be viewed as either an interpretable message-passing graph neural network or a collective classification method with higher capacity and modernized training.

Semi-supervised learning on graphs is a widely applicable problem in network science and machine learning. Two standard algorithms -- label propagation and graph neural networks -- both operate by repeatedly passing information along edges, the former by passing labels and the latter by passing node features, modulated by neural networks. These two types of algorithms have largely developed separately, and there is little understanding about the structure of network data that would make one of these approaches work particularly well compared to the other or when the approaches can be meaningfully combined. Here, we develop a Markov random field model for the data generation process of node attributes, based on correlations of attributes on and between vertices, that motivates and unifies these algorithmic approaches. We show that label propagation, a linearized graph convolutional network, and their combination can all be derived as conditional expectations under our model, when conditioning on different attributes. In addition, the data model highlights deficiencies in existing graph neural networks (while producing new algorithmic solutions), serves as a rigorous statistical framework for understanding graph learning issues such as over-smoothing, creates a testbed for evaluating inductive learning performance, and provides a way to sample graphs attributes that resemble empirical data. We also find that a new algorithm derived from our data generation model, which we call a Linear Graph Convolution, performs extremely well in practice on empirical data, and provide theoretical justification for why this is the case.

A graph neural network transforms features in each vertex's neighborhood into a vector representation of the vertex. Afterward, each vertex's representation is used independently for predicting its label. This standard pipeline implicitly assumes that vertex labels are conditionally independent given their neighborhood features. However, this is a strong assumption, and we show that it is far from true on many real-world graph datasets. Focusing on regression tasks, we find that this conditional independence assumption severely limits predictive power. This should not be that surprising, given that traditional graph-based semi-supervised learning methods such as label propagation work in the opposite fashion by explicitly modeling the correlation in predicted outcomes. Here, we address this problem with an interpretable and efficient framework that can improve any graph neural network architecture simply by exploiting correlation structure in the regression residuals. In particular, we model the joint distribution of residuals on vertices with a parameterized multivariate Gaussian, and estimate the parameters by maximizing the marginal likelihood of the observed labels. Our framework achieves substantially higher accuracy than competing baselines, and the learned parameters can be interpreted as the strength of correlation among connected vertices. Furthermore, we develop linear time algorithms for low-variance, unbiased model parameter estimates, allowing us to scale to large networks. We also provide a basic version of our method that makes stronger assumptions on correlation structure but is painless to implement, often leading to great practical performance with minimal overhead.

By including a fraction of exact exchange (EXX), hybrid functionals reduce the self-interaction error in semi-local density functional theory (DFT), and thereby furnish a more accurate and reliable description of the electronic structure in systems throughout biology, chemistry, physics, and materials science. However, the high computational cost associated with the evaluation of all required EXX quantities has limited the applicability of hybrid DFT in the treatment of large molecules and complex condensed-phase materials. To overcome this limitation, we have devised a linear-scaling yet formally exact approach that utilizes a local representation of the occupied orbitals (e.g., maximally localized Wannier functions, MLWFs) to exploit the sparsity in the real-space evaluation of the quantum mechanical exchange interaction in finite-gap systems. In this work, we present a detailed description of the theoretical and algorithmic advances required to perform MLWF-based ab initio molecular dynamics (AIMD) simulations of large-scale condensed-phase systems at the hybrid DFT level. We provide a comprehensive description of the exx algorithm, which is currently implemented in the Quantum ESPRESSO program and employs a hybrid MPI/OpenMP parallelization scheme to efficiently utilize high-performance computing (HPC) resources. This is followed by a critical assessment of the accuracy and parallel performance of this approach when performing AIMD simulations of liquid water in the canonical ensemble. With access to HPC resources, we demonstrate that exx enables hybrid DFT based AIMD simulations of condensed-phase systems containing 500-1000 atoms with a walltime cost that is comparable to semi-local DFT. In doing so, exx takes us closer to routinely performing AIMD simulations of large-scale condensed-phase systems for sufficiently long timescales at the hybrid DFT level of theory., Comment: 34 pages and 10 figures

Many time series are effectively generated by a combination of deterministic continuous flows along with discrete jumps sparked by stochastic events. However, we usually do not have the equation of motion describing the flows, or how they are affected by jumps. To this end, we introduce Neural Jump Stochastic Differential Equations that provide a data-driven approach to learn continuous and discrete dynamic behavior, i.e., hybrid systems that both flow and jump. Our approach extends the framework of Neural Ordinary Differential Equations with a stochastic process term that models discrete events. We then model temporal point processes with a piecewise-continuous latent trajectory, where the discontinuities are caused by stochastic events whose conditional intensity depends on the latent state. We demonstrate the predictive capabilities of our model on a range of synthetic and real-world marked point process datasets, including classical point processes (such as Hawkes processes), awards on Stack Overflow, medical records, and earthquake monitoring.

We present a graph-based semi-supervised learning (SSL) method for learning edge flows defined on a graph. Specifically, given flow measurements on a subset of edges, we want to predict the flows on the remaining edges. To this end, we develop a computational framework that imposes certain constraints on the overall flows, such as (approximate) flow conservation. These constraints render our approach different from classical graph-based SSL for vertex labels, which posits that tightly connected nodes share similar labels and leverages the graph structure accordingly to extrapolate from a few vertex labels to the unlabeled vertices. We derive bounds for our method's reconstruction error and demonstrate its strong performance on synthetic and real-world flow networks from transportation, physical infrastructure, and the Web. Furthermore, we provide two active learning algorithms for selecting informative edges on which to measure flow, which has applications for optimal sensor deployment. The first strategy selects edges to minimize the reconstruction error bound and works well on flows that are approximately divergence-free. The second approach clusters the graph and selects bottleneck edges that cross cluster-boundaries, which works well on flows with global trends.

Core-periphery structure is a common property of complex networks, which is a composition of tightly connected groups of core vertices and sparsely connected periphery vertices. This structure frequently emerges in traffic systems, biology, and social networks via underlying spatial positioning of the vertices. While core-periphery structure is ubiquitous, there have been limited attempts at modeling network data with this structure. Here, we develop a generative, random network model with core-periphery structure that jointly accounts for topological and spatial information by "core scores" of vertices. Our model achieves substantially higher likelihood than existing generative models of core-periphery structure, and we demonstrate how the core scores can be used in downstream data mining tasks, such as predicting airline traffic and classifying fungal networks. We also develop nearly linear time algorithms for learning model parameters and network sampling by using a method akin to the fast multipole method, a technique traditional to computational physics, which allow us to scale to networks with millions of vertices with minor tradeoffs in accuracy.

Modern x-ray sources enable the production of coherent x-ray pulses with a pulse duration in the same order as the characteristic lifetimes of core-hole states of atoms and molecules. These pulses enable the manipulation of the core-hole population during Auger decay processes, modifying the lineshape of the electron spectra. In this work, we present a theoretical model to study those effects in Neon. We identify effects in the Auger-photoelectron coincidence spectrum due to the duration and intensity of the pulses. The normal Auger line-shape is recovered in Auger electron spectrum integrated over all photoelectron energies.