Search

Large Language Models (LLMs) have shown remarkable abilities across various tasks, yet their development has predominantly centered on high-resource languages like English and Chinese, leaving low-resource languages underserved. To address this disparity, we present SeaLLMs 3, the latest iteration of the SeaLLMs model family, tailored for Southeast Asian languages. This region, characterized by its rich linguistic diversity, has lacked adequate language technology support. SeaLLMs 3 aims to bridge this gap by covering a comprehensive range of languages spoken in this region, including English, Chinese, Indonesian, Vietnamese, Thai, Tagalog, Malay, Burmese, Khmer, Lao, Tamil, and Javanese. Leveraging efficient language enhancement techniques and a specially constructed instruction tuning dataset, SeaLLMs 3 significantly reduces training costs while maintaining high performance and versatility. Our model excels in tasks such as world knowledge, mathematical reasoning, translation, and instruction following, achieving state-of-the-art performance among similarly sized models. Additionally, we prioritized safety and reliability by addressing both general and culture-specific considerations and incorporated mechanisms to reduce hallucinations. This work underscores the importance of inclusive AI, showing that advanced LLM capabilities can benefit underserved linguistic and cultural communities.

Large language models (LLMs) have demonstrated multilingual capabilities; yet, they are mostly English-centric due to the imbalanced training corpora. Existing works leverage this phenomenon to improve their multilingual performances through translation, primarily on natural language processing (NLP) tasks. This work extends the evaluation from NLP tasks to real user queries and from English-centric LLMs to non-English-centric LLMs. While translation into English can help improve the performance of multilingual NLP tasks for English-centric LLMs, it may not be optimal for all scenarios. For culture-related tasks that need deep language understanding, prompting in the native language tends to be more promising as it better captures the nuances of culture and language. Our experiments reveal varied behaviors among different LLMs and tasks in the multilingual context. Therefore, we advocate for more comprehensive multilingual evaluation and more efforts toward developing multilingual LLMs beyond English-centric ones., Comment: 19 pages

Safety of Large Language Models (LLMs) has become a critical issue given their rapid progresses. Greedy Coordinate Gradient (GCG) is shown to be effective in constructing adversarial prompts to break the aligned LLMs, but optimization of GCG is time-consuming. To reduce the time cost of GCG and enable more comprehensive studies of LLM safety, in this work, we study a new algorithm called $\texttt{Probe sampling}$. At the core of the algorithm is a mechanism that dynamically determines how similar a smaller draft model's predictions are to the target model's predictions for prompt candidates. When the target model is similar to the draft model, we rely heavily on the draft model to filter out a large number of potential prompt candidates. Probe sampling achieves up to $5.6$ times speedup using Llama2-7b-chat and leads to equal or improved attack success rate (ASR) on the AdvBench. Furthermore, probe sampling is also able to accelerate other prompt optimization techniques and adversarial methods, leading to acceleration of $1.8\times$ for AutoPrompt, $2.4\times$ for APE and $2.4\times$ for AutoDAN.

As an effective alternative to the direct fine-tuning on target tasks in specific languages, cross-lingual transfer addresses the challenges of limited training data by decoupling ''task ability'' and ''language ability'' by fine-tuning on the target task in the source language and another selected task in the target language, respectively. However, they fail to fully separate the task ability from the source language or the language ability from the chosen task. In this paper, we acknowledge the mutual reliance between task ability and language ability and direct our attention toward the gap between the target language and the source language on tasks. As the gap removes the impact of tasks, we assume that it remains consistent across tasks. Based on this assumption, we propose a new cross-lingual transfer method called $\texttt{AdaMergeX}$ that utilizes adaptive adapter merging. By introducing a reference task, we can determine that the divergence of adapters fine-tuned on the reference task in both languages follows the same distribution as the divergence of adapters fine-tuned on the target task in both languages. Hence, we can obtain target adapters by combining the other three adapters. Furthermore, we propose a structure-adaptive adapter merging method. Our empirical results demonstrate that our approach yields new and effective cross-lingual transfer, outperforming existing methods across all settings.

Large language models (LLMs) have demonstrated impressive capabilities across diverse languages. This study explores how LLMs handle multilingualism. Based on observed language ratio shifts among layers and the relationships between network structures and certain capabilities, we hypothesize the LLM's multilingual workflow ($\texttt{MWork}$): LLMs initially understand the query, converting multilingual inputs into English for task-solving. In the intermediate layers, they employ English for thinking and incorporate multilingual knowledge with self-attention and feed-forward structures, respectively. In the final layers, LLMs generate responses aligned with the original language of the query. To verify $\texttt{MWork}$, we introduce Parallel Language-specific Neuron Detection ($\texttt{PLND}$) to identify activated neurons for inputs in different languages without any labeled data. Using $\texttt{PLND}$, we validate $\texttt{MWork}$ through extensive experiments involving the deactivation of language-specific neurons across various layers and structures. Moreover, $\texttt{MWork}$ allows fine-tuning of language-specific neurons with a small dataset, enhancing multilingual abilities in a specific language without compromising others. This approach results in an average improvement of $3.6\%$ for high-resource languages and $2.3\%$ for low-resource languages across all tasks with just $400$ documents.

We propose a new method, Adversarial In-Context Learning (adv-ICL), to optimize prompt for in-context learning (ICL) by employing one LLM as a generator, another as a discriminator, and a third as a prompt modifier. As in traditional adversarial learning, adv-ICL is implemented as a two-player game between the generator and discriminator, where the generator tries to generate realistic enough output to fool the discriminator. In each round, given an input prefixed by task instructions and several exemplars, the generator produces an output. The discriminator is then tasked with classifying the generator input-output pair as model-generated or real data. Based on the discriminator loss, the prompt modifier proposes possible edits to the generator and discriminator prompts, and the edits that most improve the adversarial loss are selected. We show that adv-ICL results in significant improvements over state-of-the-art prompt optimization techniques for both open and closed-source models on 11 generation and classification tasks including summarization, arithmetic reasoning, machine translation, data-to-text generation, and the MMLU and big-bench hard benchmarks. In addition, because our method uses pre-trained models and updates only prompts rather than model parameters, it is computationally efficient, easy to extend to any LLM and task, and effective in low-resource settings., Comment: ACL 2024

Assessing factuality of text generated by large language models (LLMs) is an emerging yet crucial research area, aimed at alerting users to potential errors and guiding the development of more reliable LLMs. Nonetheless, the evaluators assessing factuality necessitate suitable evaluation themselves to gauge progress and foster advancements. This direction remains under-explored, resulting in substantial impediments to the progress of factuality evaluators. To mitigate this issue, we introduce a benchmark for Factuality Evaluation of large Language Models, referred to as felm. In this benchmark, we collect responses generated from LLMs and annotate factuality labels in a fine-grained manner. Contrary to previous studies that primarily concentrate on the factuality of world knowledge (e.g.~information from Wikipedia), felm focuses on factuality across diverse domains, spanning from world knowledge to math and reasoning. Our annotation is based on text segments, which can help pinpoint specific factual errors. The factuality annotations are further supplemented by predefined error types and reference links that either support or contradict the statement. In our experiments, we investigate the performance of several LLM-based factuality evaluators on felm, including both vanilla LLMs and those augmented with retrieval mechanisms and chain-of-thought processes. Our findings reveal that while retrieval aids factuality evaluation, current LLMs are far from satisfactory to faithfully detect factual errors., Comment: Accepted by NeurIPS 2023 Track on Datasets and Benchmarks

Breaking down a problem into intermediate steps has demonstrated impressive performance in Large Language Model (LLM) reasoning. However, the growth of the reasoning chain introduces uncertainty and error accumulation, making it challenging to elicit accurate final results. To tackle this challenge of uncertainty in multi-step reasoning, we introduce a stepwise self-evaluation mechanism to guide and calibrate the reasoning process of LLMs. We propose a decoding algorithm integrating the self-evaluation guidance via stochastic beam search. The self-evaluation guidance serves as a better-calibrated automatic criterion, facilitating an efficient search in the reasoning space and resulting in superior prediction quality. Stochastic beam search balances exploitation and exploration of the search space with temperature-controlled randomness. Our approach surpasses the corresponding Codex-backboned baselines in few-shot accuracy by $6.34\%$, $9.56\%$, and $5.46\%$ on the GSM8K, AQuA, and StrategyQA benchmarks, respectively. Experiment results with Llama-2 on arithmetic reasoning demonstrate the efficiency of our method in outperforming the baseline methods with comparable computational budgets. Further analysis in multi-step reasoning finds our self-evaluation guidance pinpoints logic failures and leads to higher consistency and robustness. Our code is publicly available at https://guideddecoding.github.io/., Comment: NeurIPS 2023. 10 pages, 7 figures, 4 tables (33 pages, 14 figures, 15 tables including references and appendices)

We consider sequential state and parameter learning in state-space models with intractable state transition and observation processes. By exploiting low-rank tensor train (TT) decompositions, we propose new sequential learning methods for joint parameter and state estimation under the Bayesian framework. Our key innovation is the introduction of scalable function approximation tools such as TT for recursively learning the sequentially updated posterior distributions. The function approximation perspective of our methods offers tractable error analysis and potentially alleviates the particle degeneracy faced by many particle-based methods. In addition to the new insights into the algorithmic design, our methods complement conventional particle-based methods. Our TT-based approximations naturally define conditional Knothe--Rosenblatt (KR) rearrangements that lead to parameter estimation, filtering, smoothing and path estimation accompanying our sequential learning algorithms, which open the door to removing potential approximation bias. We also explore several preconditioning techniques based on either linear or nonlinear KR rearrangements to enhance the approximation power of TT for practical problems. We demonstrate the efficacy and efficiency of our proposed methods on several state-space models, in which our methods achieve state-of-the-art estimation accuracy and computational performance.

Broadband dual-comb spectroscopy has attracted increasing interests due to its unique advantages in high spectral resolution, fast detection, and so on. Although the dual-comb technique is relatively mature in the infrared wavelengths, it is, currently, not commercially capable of practical applications in the terahertz regime due to the lack of high performance broadband terahertz dual-comb sources. In the terahertz frequency range, the electrically pumped quantum cascade laser (QCL) is a suitable candidate for the dual-comb operation. However, free running terahertz QCL dual-comb sources normally show limited optical bandwidths ($\sim$100-200 GHz). Although the resonant microwave injection locking has been widely used to broaden the emission spectra of terahertz QCLs by modulating the laser drive current at the cavity round-trip frequency, it is hard to be employed to broaden the dual-comb bandwidths due to the large phase noise induced by the resonant injection and non-ideal microwave circuits. Therefore, it is challenging to obtain broadband terahertz dual-comb sources that can fully exploits the laser gain bandwidth. Here, we employ an off-resonant microwave injection to significantly broaden the dual-comb bandwidth of a terahertz QCL dual-comb source emitting around 4.2 THz. The measured optical dual-comb bandwidth is broadened from 147 GHz in free running to $>$450 GHz under the off-resonant injection. The broadened dual-comb bandwidth is experimentally proved by the transmission measurements of a filter and a GaAs etalon. By performing a simple numerical analysis based on a rate equation model, we explain that the broadband dual-comb operation under the off-resonant microwave injection could be resulted from a wider lasing bandwidth and a higher degree of phase matching., Comment: 22 pages, 6 figures

The paper presents an efficient real-time scheduling algorithm for intelligent real-time edge services, defined as those that perform machine intelligence tasks, such as voice recognition, LIDAR processing, or machine vision, on behalf of local embedded devices that are themselves unable to support extensive computations. The work contributes to a recent direction in real-time computing that develops scheduling algorithms for machine intelligence tasks with anytime prediction. We show that deep neural network workflows can be cast as imprecise computations, each with a mandatory part and (several) optional parts whose execution utility depends on input data. The goal of the real-time scheduler is to maximize the average accuracy of deep neural network outputs while meeting task deadlines, thanks to opportunistic shedding of the least necessary optional parts. The work is motivated by the proliferation of increasingly ubiquitous but resource-constrained embedded devices (for applications ranging from autonomous cars to the Internet of Things) and the desire to develop services that endow them with intelligence. Experiments on recent GPU hardware and a state of the art deep neural network for machine vision illustrate that our scheme can increase the overall accuracy by 10%-20% while incurring (nearly) no deadline misses.

Dual-comb sources with equally spaced and low phase noise frequency lines are of great importance for high resolution spectroscopy and metrology. In the terahertz frequency range, electrically pumped semiconductor quantum cascade lasers (QCLs) are suitable candidates for frequency comb and dual-comb operation. For a single laser frequency comb, the repetition rate can be locked using a microwave injection locking and the carrier frequency can be locked to a highly stable source. However, for the locking of two laser combs, four frequencies (two repetition rates and two carrier offset frequencies) should be simultaneously locked; If one only refers to the dual-comb signal, two relative frequencies, i.e., the offset frequency and repetition frequency of one laser against those of the other laser, should be locked. Although the locking techniques that have been successfully used for a single laser comb can be, in principle, applied to a dual-comb laser source, the complete locking considerably complicates the implementation of such a system. Here, we propose a method to stabilize a terahertz QCL dual-comb source by phase locking one of the dual-comb lines to a radio frequency (RF) synthesizer. This technique forces one of the lasers to follow the tone of the other one (keeping the sum of the carrier offset frequency difference and repetition frequency difference between the two laser combs as a constant) by exploiting a laser self-detection that avoids the use of an external detector. Through the demonstration of this locking technique, we demonstrate that the dual-comb can generate periodic pulses over a 2 us time scale, showing that the terahertz QCL comb without a control of the repetition rate can produce pulsed-type waveforms., Comment: 21 pages, 5 figures

Recent advances in deep learning motivate the use of deep neural networks in Internet-of-Things (IoT) applications. These networks are modelled after signal processing in the human brain, thereby leading to significant advantages at perceptual tasks such as vision and speech recognition. IoT applications, however, often measure physical phenomena, where the underlying physics (such as inertia, wireless signal propagation, or the natural frequency of oscillation) are fundamentally a function of signal frequencies, offering better features in the frequency domain. This observation leads to a fundamental question: For IoT applications, can one develop a new brand of neural network structures that synthesize features inspired not only by the biology of human perception but also by the fundamental nature of physics? Hence, in this paper, instead of using conventional building blocks (e.g., convolutional and recurrent layers), we propose a new foundational neural network building block, the Short-Time Fourier Neural Network (STFNet). It integrates a widely-used time-frequency analysis method, the Short-Time Fourier Transform, into data processing to learn features directly in the frequency domain, where the physics of underlying phenomena leave better foot-prints. STFNets bring additional flexibility to time-frequency analysis by offering novel nonlinear learnable operations that are spectral-compatible. Moreover, STFNets show that transforming signals to a domain that is more connected to the underlying physics greatly simplifies the learning process. We demonstrate the effectiveness of STFNets with extensive experiments. STFNets significantly outperform the state-of-the-art deep learning models in all experiments. A STFNet, therefore, demonstrates superior capability as the fundamental building block of deep neural networks for IoT applications for various sensor inputs.

Deep neural networks show great potential as solutions to many sensing application problems, but their excessive resource demand slows down execution time, pausing a serious impediment to deployment on low-end devices. To address this challenge, recent literature focused on compressing neural network size to improve performance. We show that changing neural network size does not proportionally affect performance attributes of interest, such as execution time. Rather, extreme run-time nonlinearities exist over the network configuration space. Hence, we propose a novel framework, called FastDeepIoT, that uncovers the non-linear relation between neural network structure and execution time, then exploits that understanding to find network configurations that significantly improve the trade-off between execution time and accuracy on mobile and embedded devices. FastDeepIoT makes two key contributions. First, FastDeepIoT automatically learns an accurate and highly interpretable execution time model for deep neural networks on the target device. This is done without prior knowledge of either the hardware specifications or the detailed implementation of the used deep learning library. Second, FastDeepIoT informs a compression algorithm how to minimize execution time on the profiled device without impacting accuracy. We evaluate FastDeepIoT using three different sensing-related tasks on two mobile devices: Nexus 5 and Galaxy Nexus. FastDeepIoT further reduces the neural network execution time by $48\%$ to $78\%$ and energy consumption by $37\%$ to $69\%$ compared with the state-of-the-art compression algorithms., Comment: Accepted by SenSys '18