Quantization techniques are widely used to improve inference speed and deployment of large language models. While a wide body of work examines the impact of quantized LLMs on English tasks, none have examined the effect of quantization across languages. We conduct a thorough analysis of quantized multilingual LLMs, focusing on their performance across languages and at varying scales. We use automatic benchmarks, LLM-as-a-Judge methods, and human evaluation, finding that (1) harmful effects of quantization are apparent in human evaluation, and automatic metrics severely underestimate the detriment: a 1.7% average drop in Japanese across automatic tasks corresponds to a 16.0% drop reported by human evaluators on realistic prompts; (2) languages are disparately affected by quantization, with non-Latin script languages impacted worst; and (3) challenging tasks such as mathematical reasoning degrade fastest. As the ability to serve low-compute models is critical for wide global adoption of NLP technologies, our results urge consideration of multilingual performance as a key evaluation criterion for efficient models.

While federated learning leverages distributed client resources, it faces challenges due to heterogeneous client capabilities. This necessitates allocating models suited to clients' resources and careful parameter aggregation to accommodate this heterogeneity. We propose HypeMeFed, a novel federated learning framework for supporting client heterogeneity by combining a multi-exit network architecture with hypernetwork-based model weight generation. This approach aligns the feature spaces of heterogeneous model layers and resolves per-layer information disparity during weight aggregation. To practically realize HypeMeFed, we also propose a low-rank factorization approach to minimize computation and memory overhead associated with hypernetworks. Our evaluations on a real-world heterogeneous device testbed indicate that HypeMeFed enhances accuracy by 5.12% over FedAvg, reduces the hypernetwork memory requirements by 98.22%, and accelerates its operations by 1.86 times compared to a naive hypernetwork approach. These results demonstrate HypeMeFed's effectiveness in leveraging and engaging heterogeneous clients for federated learning.

Requiring a Large Language Model to generate intermediary reasoning steps has been shown to be an effective way of boosting performance. In fact, it has been found that instruction tuning on these intermediary reasoning steps improves model performance. In this work, we present a novel method of further improving performance by requiring models to compare multiple reasoning chains before generating a solution in a single inference step. We call this method Divergent CoT (DCoT). We find that instruction tuning on DCoT datasets boosts the performance of even smaller, and therefore more accessible, LLMs. Through a rigorous set of experiments spanning a wide range of tasks that require various reasoning types, we show that fine-tuning on DCoT consistently improves performance over the CoT baseline across model families and scales (1.3B to 70B). Through a combination of empirical and manual evaluation, we additionally show that these performance gains stem from models generating multiple divergent reasoning chains in a single inference step, indicative of the enabling of self-correction in language models. Our code and data are publicly available at https://github.com/UKPLab/arxiv2024-divergent-cot.

In the age of big data and advanced computational methods, the prediction of large-scale social behaviors, reminiscent of Isaac Asimov's fictional science of Psychohistory, is becoming increasingly feasible. This paper consists of a theoretical exploration of the integration of computational power and mathematical frameworks, particularly through Topological Data Analysis (TDA) (Carlsson, Vejdemo-Johansson, 2022) and Artificial Intelligence (AI), to forecast societal trends through social media data analysis. By examining social media as a reflective surface of collective human behavior through the systematic behaviorist approach (Glenn, et al., 2016), I argue that these tools provide unprecedented clarity into the dynamics of large communities. This study dialogues with Asimov's work, drawing parallels between his visionary concepts and contemporary methodologies, illustrating how modern computational techniques can uncover patterns and predict shifts in social behavior, contributing to the emerging field of digital sociology -- or even, Psychohistory itself.

With the advent of large models based on the Transformer architecture, researchers have observed an anomalous phenomenon in the Attention mechanism--there is a very high attention on the first element, which is prevalent across Transformer-based models. It is crucial to understand it for the development of techniques focusing on attention distribution, such as Key-Value (KV) Cache compression and infinite extrapolation; however, the latent cause leaves to be unknown. In this paper, we analyze such a phenomenon from the perspective of waiver phenomenon, which involves reducing the internal values of certain elements in the sequence, allowing them to absorb excess attention without affecting their contribution to information. In specific models, due to differences in positional encoding and attention patterns, we have found that the selection of waiver elements by the model can be categorized into two methods: positional-encoding-based and feature-distribution-within-elements-based.

Many recent works have explored using language models for planning problems. One line of research focuses on translating natural language descriptions of planning tasks into structured planning languages, such as the planning domain definition language (PDDL). While this approach is promising, accurately measuring the quality of generated PDDL code continues to pose significant challenges. First, generated PDDL code is typically evaluated using planning validators that check whether the problem can be solved with a planner. This method is insufficient because a language model might generate valid PDDL code that does not align with the natural language description of the task. Second, existing evaluation sets often have natural language descriptions of the planning task that closely resemble the ground truth PDDL, reducing the challenge of the task. To bridge this gap, we introduce \benchmarkName, a benchmark designed to evaluate language models' ability to generate PDDL code from natural language descriptions of planning tasks. We begin by creating a PDDL equivalence algorithm that rigorously evaluates the correctness of PDDL code generated by language models by flexibly comparing it against a ground truth PDDL. Then, we present a dataset of $132,037$ text-to-PDDL pairs across 13 different tasks, with varying levels of difficulty. Finally, we evaluate several API-access and open-weight language models that reveal this task's complexity. For example, $87.6\%$ of the PDDL problem descriptions generated by GPT-4o are syntactically parseable, $82.2\%$ are valid, solve-able problems, but only $35.1\%$ are semantically correct, highlighting the need for a more rigorous benchmark for this problem.

We introduce Blink, a new benchmark for multimodal language models (LLMs) that focuses on core visual perception abilities not found in other evaluations. Most of the Blink tasks can be solved by humans "within a blink" (e.g., relative depth estimation, visual correspondence, forensics detection, and multi-view reasoning). However, we find these perception-demanding tasks cast significant challenges for current multimodal LLMs because they resist mediation through natural language. Blink reformats 14 classic computer vision tasks into 3,807 multiple-choice questions, paired with single or multiple images and visual prompting. While humans get 95.70% accuracy on average, Blink is surprisingly challenging for existing multimodal LLMs: even the best-performing GPT-4V and Gemini achieve accuracies of 51.26% and 45.72%, only 13.17% and 7.63% higher than random guessing, indicating that such perception abilities have not "emerged" yet in recent multimodal LLMs. Our analysis also highlights that specialist CV models could solve these problems much better, suggesting potential pathways for future improvements. We believe Blink will stimulate the community to help multimodal LLMs catch up with human-level visual perception.

Maximizing the likelihood of the next token is an established, statistically sound objective for pre-training language models. In this paper we show that we can train better models faster by pre-aggregating the corpus with a collapsed $n$-gram distribution. Previous studies have proposed corpus-level $n$-gram statistics as a regularizer; however, the construction and querying of such $n$-grams, if done naively, prove to be costly and significantly impede training speed, thereby limiting their application in modern large language model pre-training. We introduce an alternative compact representation of the next token distribution that, in expectation, aligns with the complete $n$-gram distribution while markedly reducing variance across mini-batches compared to the standard next-token loss. Empirically, we demonstrate that both the $n$-gram regularized model and our approximation yield substantial improvements in model quality and convergence rate compared to existing methods. Furthermore, our approximation facilitates scalability of gains to larger datasets and models compared to the straightforward $n$-gram regularization method.

Proton resonance frequency (PRF) based MR thermometry is essential for focused ultrasound (FUS) thermal ablation therapies. This work aims to enhance temporal resolution in dynamic MR temperature map reconstruction using an improved deep learning method. The training-optimized methods and five classical neural networks were applied on the 2-fold and 4-fold under-sampling k-space data to reconstruct the temperature maps. The enhanced training modules included offline/online data augmentations, knowledge distillation, and the amplitude-phase decoupling loss function. The heating experiments were performed by a FUS transducer on phantom and ex vivo tissues, respectively. These data were manually under-sampled to imitate acceleration procedures and trained in our method to get the reconstruction model. The additional dozen or so testing datasets were separately obtained for evaluating the real-time performance and temperature accuracy. Acceleration factors of 1.9 and 3.7 were found for 2 times and 4 times k-space under-sampling strategies and the ResUNet-based deep learning reconstruction performed exceptionally well. In 2-fold acceleration scenario, the RMSE of temperature map patches provided the values of 0.888 degree centigrade and 1.145 degree centigrade on phantom and ex vivo testing datasets. The DICE value of temperature areas enclosed by 43 degree centigrade isotherm was 0.809, and the Bland-Altman analysis showed a bias of -0.253 degree centigrade with the apart of plus or minus 2.16 degree centigrade. In 4 times under-sampling case, these evaluating values decreased by approximately 10%. This study demonstrates that deep learning-based reconstruction can significantly enhance the accuracy and efficiency of MR thermometry for clinical FUS thermal therapies.

This paper introduces a novel combination of two tasks, previously treated separately: acoustic-to-articulatory speech inversion (AAI) and phoneme-to-articulatory (PTA) motion estimation. We refer to this joint task as acoustic phoneme-to-articulatory speech inversion (APTAI) and explore two different approaches, both working speaker- and text-independently during inference. We use a multi-task learning setup, with the end-to-end goal of taking raw speech as input and estimating the corresponding articulatory movements, phoneme sequence, and phoneme alignment. While both proposed approaches share these same requirements, they differ in their way of achieving phoneme-related predictions: one is based on frame classification, the other on a two-staged training procedure and forced alignment. We reach competitive performance of 0.73 mean correlation for the AAI task and achieve up to approximately 87% frame overlap compared to a state-of-the-art text-dependent phoneme force aligner.