Patrick Huber

Hanxian Huang, Igor Fedorov, Andrey Gromov et al. · meta-ai, mila

Real-time AI experiences call for on-device large language models (OD-LLMs) optimized for efficient deployment on resource-constrained hardware. The most useful OD-LLMs produce near-real-time responses and exhibit broad hardware compatibility, maximizing user reach. We present a methodology for designing such models using hardware-in-the-loop architecture search under mobile latency constraints. This system is amenable to industry-scale deployment: it generates models deployable without custom kernels and compatible with standard mobile runtimes like Executorch. Our methodology avoids specialized attention mechanisms and instead uses attention skipping for long-context acceleration. Our approach jointly optimizes model architecture (layers, dimensions) and attention pattern. To efficiently evaluate candidates, we treat each as a pruned version of a pretrained backbone with inherited weights, thereby achieving high accuracy with minimal continued pretraining. We leverage the low cost of latency evaluation in a staged process: learning an accurate latency model first, then searching for the Pareto-frontier across latency and quality. This yields MobileLLM-Flash, a family of foundation models (350M, 650M, 1.4B) for efficient on-device use with strong capabilities, supporting up to 8k context length. MobileLLM-Flash delivers up to 1.8x and 1.6x faster prefill and decode on mobile CPUs with comparable or superior quality. Our analysis of Pareto-frontier design choices offers actionable principles for OD-LLM design.

Chuyuan Li, Patrick Huber, Wen Xiao et al.

Discourse processing suffers from data sparsity, especially for dialogues. As a result, we explore approaches to build discourse structures for dialogues, based on attention matrices from Pre-trained Language Models (PLMs). We investigate multiple tasks for fine-tuning and show that the dialogue-tailored Sentence Ordering task performs best. To locate and exploit discourse information in PLMs, we propose an unsupervised and a semi-supervised method. Our proposals achieve encouraging results on the STAC corpus, with F1 scores of 57.2 and 59.3 for unsupervised and semi-supervised methods, respectively. When restricted to projective trees, our scores improved to 63.3 and 68.1.

Patrick Huber, Giuseppe Carenini

With a growing number of BERTology work analyzing different components of pre-trained language models, we extend this line of research through an in-depth analysis of discourse information in pre-trained and fine-tuned language models. We move beyond prior work along three dimensions: First, we describe a novel approach to infer discourse structures from arbitrarily long documents. Second, we propose a new type of analysis to explore where and how accurately intrinsic discourse is captured in the BERT and BART models. Finally, we assess how similar the generated structures are to a variety of baselines as well as their distribution within and between models.

Linzi Xing, Patrick Huber, Giuseppe Carenini

Recent neural supervised topic segmentation models achieve distinguished superior effectiveness over unsupervised methods, with the availability of large-scale training corpora sampled from Wikipedia. These models may, however, suffer from limited robustness and transferability caused by exploiting simple linguistic cues for prediction, but overlooking more important inter-sentential topical consistency. To address this issue, we present a discourse-aware neural topic segmentation model with the injection of above-sentence discourse dependency structures to encourage the model make topic boundary prediction based more on the topical consistency between sentences. Our empirical study on English evaluation datasets shows that injecting above-sentence discourse structures to a neural topic segmenter with our proposed strategy can substantially improve its performances on intra-domain and out-of-domain data, with little increase of model's complexity.

Patrick Huber, Arash Einolghozati, Rylan Conway et al. · meta-ai

Distilling conversational skills into Small Language Models (SLMs) with approximately 1 billion parameters presents significant challenges. Firstly, SLMs have limited capacity in their model parameters to learn extensive knowledge compared to larger models. Secondly, high-quality conversational datasets are often scarce, small, and domain-specific. Addressing these challenges, we introduce a novel data distillation framework named CoDi (short for Conversational Distillation, pronounced "Cody"), allowing us to synthesize large-scale, assistant-style datasets in a steerable and diverse manner. Specifically, while our framework is task agnostic at its core, we explore and evaluate the potential of CoDi on the task of conversational grounded reasoning for question answering. This is a typical on-device scenario for specialist SLMs, allowing for open-domain model responses, without requiring the model to "memorize" world knowledge in its limited weights. Our evaluations show that SLMs trained with CoDi-synthesized data achieve performance comparable to models trained on human-annotated data in standard metrics. Additionally, when using our framework to generate larger datasets from web data, our models surpass larger, instruction-tuned models in zero-shot conversational grounded reasoning tasks.

Patrick Huber, Giuseppe Carenini

Discourse analysis and discourse parsing have shown great impact on many important problems in the field of Natural Language Processing (NLP). Given the direct impact of discourse annotations on model performance and interpretability, robustly extracting discourse structures from arbitrary documents is a key task to further improve computational models in NLP. To this end, we present a new, supervised paradigm directly tackling the domain adaptation issue in discourse parsing. Specifically, we introduce the first fully supervised discourse parser designed to alleviate the domain dependency through a staged model of weak classifiers by introducing the gradient boosting framework.

Patrick Huber, Giuseppe Carenini

Discourse parsing is an essential upstream task in Natural Language Processing with strong implications for many real-world applications. Despite its widely recognized role, most recent discourse parsers (and consequently downstream tasks) still rely on small-scale human-annotated discourse treebanks, trying to infer general-purpose discourse structures from very limited data in a few narrow domains. To overcome this dire situation and allow discourse parsers to be trained on larger, more diverse and domain-independent datasets, we propose a framework to generate "silver-standard" discourse trees from distant supervision on the auxiliary task of sentiment analysis.

Patrick Huber, Giuseppe Carenini

With a growing need for robust and general discourse structures in many downstream tasks and real-world applications, the current lack of high-quality, high-quantity discourse trees poses a severe shortcoming. In order the alleviate this limitation, we propose a new strategy to generate tree structures in a task-agnostic, unsupervised fashion by extending a latent tree induction framework with an auto-encoding objective. The proposed approach can be applied to any tree-structured objective, such as syntactic parsing, discourse parsing and others. However, due to the especially difficult annotation process to generate discourse trees, we initially develop such method to complement task-specific models in generating much larger and more diverse discourse treebanks.

Yuxiao Qu, Jinmeng Rao, Song Gao et al.

Understanding and measuring the resilience of food supply networks is a global imperative to tackle increasing food insecurity. However, the complexity of these networks, with their multidimensional interactions and decisions, presents significant challenges. This paper proposes FLEE-GNN, a novel Federated Learning System for Edge-Enhanced Graph Neural Network, designed to overcome these challenges and enhance the analysis of geospatial resilience of multicommodity food flow network, which is one type of spatial networks. FLEE-GNN addresses the limitations of current methodologies, such as entropy-based methods, in terms of generalizability, scalability, and data privacy. It combines the robustness and adaptability of graph neural networks with the privacy-conscious and decentralized aspects of federated learning on food supply network resilience analysis across geographical regions. This paper also discusses FLEE-GNN's innovative data generation techniques, experimental designs, and future directions for improvement. The results show the advancements of this approach to quantifying the resilience of multicommodity food flow networks, contributing to efforts towards ensuring global food security using AI methods. The developed FLEE-GNN has the potential to be applied in other spatial networks with spatially heterogeneous sub-network distributions.

Zekun Li, Zhiyu Zoey Chen, Mike Ross et al. · microsoft-research

Large language models (LLMs) are increasingly prevalent in conversational systems due to their advanced understanding and generative capabilities in general contexts. However, their effectiveness in task-oriented dialogues (TOD), which requires not only response generation but also effective dialogue state tracking (DST) within specific tasks and domains, remains less satisfying. In this work, we propose a novel approach FnCTOD for solving DST with LLMs through function calling. This method improves zero-shot DST, allowing adaptation to diverse domains without extensive data collection or model tuning. Our experimental results demonstrate that our approach achieves exceptional performance with both modestly sized open-source and also proprietary LLMs: with in-context prompting it enables various 7B or 13B parameter models to surpass the previous state-of-the-art (SOTA) achieved by ChatGPT, and improves ChatGPT's performance beating the SOTA by 5.6% average joint goal accuracy (JGA). Individual model results for GPT-3.5 and GPT-4 are boosted by 4.8% and 14%, respectively. We also show that by fine-tuning on a small collection of diverse task-oriented dialogues, we can equip modestly sized models, specifically a 13B parameter LLaMA2-Chat model, with function-calling capabilities and DST performance comparable to ChatGPT while maintaining their chat capabilities. We have made the code publicly available at https://github.com/facebookresearch/FnCTOD

Patrick Huber, Ernie Chang, Wei Wen et al.

Efficient on-device language models around 1 billion parameters are essential for powering low-latency AI applications on mobile and wearable devices. However, achieving strong performance in this model class, while supporting long context windows and practical deployment remains a significant challenge. We introduce MobileLLM-Pro, a 1-billion-parameter language model optimized for on-device deployment. MobileLLM-Pro achieves state-of-the-art results across 11 standard benchmarks, significantly outperforming both Gemma 3-1B and Llama 3.2-1B, while supporting context windows of up to 128,000 tokens and showing only minor performance regressions at 4-bit quantization. These improvements are enabled by four core innovations: (1) implicit positional distillation, a novel technique that effectively instills long-context capabilities through knowledge distillation; (2) a specialist model merging framework that fuses multiple domain experts into a compact model without parameter growth; (3) simulation-driven data mixing using utility estimation; and (4) 4-bit quantization-aware training with self-distillation. We release our model weights and code to support future research in efficient on-device language models.

Sahithya Ravi, Patrick Huber, Akshat Shrivastava et al. · meta-ai

The emergence of Large Language Models (LLMs) has brought to light promising language generation capabilities, particularly in performing tasks like complex reasoning and creative writing. Consequently, distillation through imitation of teacher responses has emerged as a popular technique to transfer knowledge from LLMs to more accessible, Small Language Models (SLMs). While this works well for simpler tasks, there is a substantial performance gap on tasks requiring intricate language comprehension and creativity, such as humor generation. We hypothesize that this gap may stem from the fact that creative tasks might be hard to learn by imitation alone and explore whether an approach, involving supplementary guidance from the teacher, could yield higher performance. To address this, we study the effect of assigning a dual role to the LLM - as a "teacher" generating data, as well as a "critic" evaluating the student's performance. Our experiments on humor generation reveal that the incorporation of feedback significantly narrows the performance gap between SLMs and their larger counterparts compared to merely relying on imitation. As a result, our research highlights the potential of using feedback as an additional dimension to data when transferring complex language abilities via distillation.

Ernie Chang, Yang Li, Patrick Huber et al.

In language model training, it is desirable to equip models with capabilities from various tasks. However, it is not clear how to directly obtain the right data mixtures for these capabilities as the relationship between data and tasks is difficult to be modeled. In this work, we observe that checkpoint models exhibit emerging capabilities at different points in the training trajectory. Often, the training process saves checkpoints as artifacts that are under-utilized as a source of in-training data signals. We identify these artifact models based on their respective capabilities on the benchmarks and leverage them as data mixers by using their aggregated first-order influence approximation over source data. We demonstrated on eight reasoning benchmarks that the proposed framework shows significant improvements in the pretraining setting, with performance improvements of up to 1.93%. Overall, this shows the potential of checkpoint models to enhance data quality and optimize data mixtures.

Patrick Huber, Ernie Chang, Chinnadhurai Sankar et al.

Extending the context window of language models typically requires expensive long-context pre-training, posing significant challenges for both training efficiency and data collection. In this paper, we present evidence that long-context retrieval capabilities can be transferred to student models through logit-based knowledge distillation, even when training exclusively on packed short-context samples within a long-context window. We provide comprehensive insights through the lens of Rotary Position Embedding (RoPE) and establish three key findings. First, consistent with prior work, we show that phase-wise RoPE scaling, which maximizes rotational spectrum utilization at each training stage, also achieves the best long-context performance in knowledge distillation setups. Second, we demonstrate that logit-based knowledge distillation can directly enable positional information transfer. Using an experimental setup with packed repeated token sequences, we trace the propagation of positional perturbations from query and key vectors through successive transformer layers to output logits, revealing that positional information systematically influences the teacher's output distribution and, in turn, the distillation signal received by the student model. Third, our analysis uncovers structured update patterns in the query state during long-context extension, with distinct parameter spans exhibiting strong sensitivity to long-context training.

Patrick Huber, Akshat Shrivastava, Ernie Chang et al. · meta-ai, mila

Sparse Mixture of Expert (MoE) models are popular foundational architectures at large scale, however, under-explored at smaller sizes. Here, we show how to enable Compact Sparse Mixture of Experts (CoSMoEs) for on-device inference. Specifically, we tackle the three main on-device dimensions: Quality, Memory and Latency. Along the quality axis, we show that in a fair evaluation (removing confounding factors) MoE architectures outperform FLOP-aligned dense models at on-device scale. We introduce weight-decomposed experts, further improving the MoE model performance. Regarding model memory and latency, we significantly improve model offloading efficiency and, in turn, reduce model inference latency.

Patrick Huber, Linzi Xing, Giuseppe Carenini

RST-style discourse parsing plays a vital role in many NLP tasks, revealing the underlying semantic/pragmatic structure of potentially complex and diverse documents. Despite its importance, one of the most prevailing limitations in modern day discourse parsing is the lack of large-scale datasets. To overcome the data sparsity issue, distantly supervised approaches from tasks like sentiment analysis and summarization have been recently proposed. Here, we extend this line of research by exploiting distant supervision from topic segmentation, which can arguably provide a strong and oftentimes complementary signal for high-level discourse structures. Experiments on two human-annotated discourse treebanks confirm that our proposal generates accurate tree structures on sentence and paragraph level, consistently outperforming previous distantly supervised models on the sentence-to-document task and occasionally reaching even higher scores on the sentence-to-paragraph level.

Patrick Huber, Armen Aghajanyan, Barlas Oğuz et al.

With the rise of large-scale pre-trained language models, open-domain question-answering (ODQA) has become an important research topic in NLP. Based on the popular pre-training fine-tuning approach, we posit that an additional in-domain pre-training stage using a large-scale, natural, and diverse question-answering (QA) dataset can be beneficial for ODQA. Consequently, we propose a novel QA dataset based on the Common Crawl project in this paper. Using the readily available schema.org annotation, we extract around 130 million multilingual question-answer pairs, including about 60 million English data-points. With this previously unseen number of natural QA pairs, we pre-train popular language models to show the potential of large-scale in-domain pre-training for the task of question-answering. In our experiments, we find that pre-training question-answering models on our Common Crawl Question Answering dataset (CCQA) achieves promising results in zero-shot, low resource and fine-tuned settings across multiple tasks, models and benchmarks.

Patrick Huber, Wen Xiao, Giuseppe Carenini

Aiming for a better integration of data-driven and linguistically-inspired approaches, we explore whether RST Nuclearity, assigning a binary assessment of importance between text segments, can be replaced by automatically generated, real-valued scores, in what we call a Weighted-RST framework. In particular, we find that weighted discourse trees from auxiliary tasks can benefit key NLP downstream applications, compared to nuclearity-centered approaches. We further show that real-valued importance distributions partially and interestingly align with the assessment and uncertainty of human annotators.

Wen Xiao, Patrick Huber, Giuseppe Carenini

Previous work indicates that discourse information benefits summarization. In this paper, we explore whether this synergy between discourse and summarization is bidirectional, by inferring document-level discourse trees from pre-trained neural summarizers. In particular, we generate unlabeled RST-style discourse trees from the self-attention matrices of the transformer model. Experiments across models and datasets reveal that the summarizer learns both, dependency- and constituency-style discourse information, which is typically encoded in a single head, covering long- and short-distance discourse dependencies. Overall, the experimental results suggest that the learned discourse information is general and transferable inter-domain.

Patrick Huber, Giuseppe Carenini

Discourse information, as postulated by popular discourse theories, such as RST and PDTB, has been shown to improve an increasing number of downstream NLP tasks, showing positive effects and synergies of discourse with important real-world applications. While methods for incorporating discourse become more and more sophisticated, the growing need for robust and general discourse structures has not been sufficiently met by current discourse parsers, usually trained on small scale datasets in a strictly limited number of domains. This makes the prediction for arbitrary tasks noisy and unreliable. The overall resulting lack of high-quality, high-quantity discourse trees poses a severe limitation to further progress. In order the alleviate this shortcoming, we propose a new strategy to generate tree structures in a task-agnostic, unsupervised fashion by extending a latent tree induction framework with an auto-encoding objective. The proposed approach can be applied to any tree-structured objective, such as syntactic parsing, discourse parsing and others. However, due to the especially difficult annotation process to generate discourse trees, we initially develop a method to generate larger and more diverse discourse treebanks. In this paper we are inferring general tree structures of natural text in multiple domains, showing promising results on a diverse set of tasks.

Wen Xiao, Patrick Huber, Giuseppe Carenini

The multi-head self-attention of popular transformer models is widely used within Natural Language Processing (NLP), including for the task of extractive summarization. With the goal of analyzing and pruning the parameter-heavy self-attention mechanism, there are multiple approaches proposing more parameter-light self-attention alternatives. In this paper, we present a novel parameter-lean self-attention mechanism using discourse priors. Our new tree self-attention is based on document-level discourse information, extending the recently proposed "Synthesizer" framework with another lightweight alternative. We show empirical results that our tree self-attention approach achieves competitive ROUGE-scores on the task of extractive summarization. When compared to the original single-head transformer model, the tree attention approach reaches similar performance on both, EDU and sentence level, despite the significant reduction of parameters in the attention component. We further significantly outperform the 8-head transformer model on sentence level when applying a more balanced hyper-parameter setting, requiring an order of magnitude less parameters.

Grigorii Guz, Patrick Huber, Giuseppe Carenini

RST-based discourse parsing is an important NLP task with numerous downstream applications, such as summarization, machine translation and opinion mining. In this paper, we demonstrate a simple, yet highly accurate discourse parser, incorporating recent contextual language models. Our parser establishes the new state-of-the-art (SOTA) performance for predicting structure and nuclearity on two key RST datasets, RST-DT and Instr-DT. We further demonstrate that pretraining our parser on the recently available large-scale "silver-standard" discourse treebank MEGA-DT provides even larger performance benefits, suggesting a novel and promising research direction in the field of discourse analysis.

Patrick Huber, Giuseppe Carenini

Sentiment analysis, especially for long documents, plausibly requires methods capturing complex linguistics structures. To accommodate this, we propose a novel framework to exploit task-related discourse for the task of sentiment analysis. More specifically, we are combining the large-scale, sentiment-dependent MEGA-DT treebank with a novel neural architecture for sentiment prediction, based on a hybrid TreeLSTM hierarchical attention model. Experiments show that our framework using sentiment-related discourse augmentations for sentiment prediction enhances the overall performance for long documents, even beyond previous approaches using well-established discourse parsers trained on human annotated data. We show that a simple ensemble approach can further enhance performance by selectively using discourse, depending on the document length.

Patrick Huber, Giuseppe Carenini

The lack of large and diverse discourse treebanks hinders the application of data-driven approaches, such as deep-learning, to RST-style discourse parsing. In this work, we present a novel scalable methodology to automatically generate discourse treebanks using distant supervision from sentiment-annotated datasets, creating and publishing MEGA-DT, a new large-scale discourse-annotated corpus. Our approach generates discourse trees incorporating structure and nuclearity for documents of arbitrary length by relying on an efficient heuristic beam-search strategy, extended with a stochastic component. Experiments on multiple datasets indicate that a discourse parser trained on our MEGA-DT treebank delivers promising inter-domain performance gains when compared to parsers trained on human-annotated discourse corpora.

Patrick Huber, Giuseppe Carenini

Discourse parsing could not yet take full advantage of the neural NLP revolution, mostly due to the lack of annotated datasets. We propose a novel approach that uses distant supervision on an auxiliary task (sentiment classification), to generate abundant data for RST-style discourse structure prediction. Our approach combines a neural variant of multiple-instance learning, using document-level supervision, with an optimal CKY-style tree generation algorithm. In a series of experiments, we train a discourse parser (for only structure prediction) on our automatically generated dataset and compare it with parsers trained on human-annotated corpora (news domain RST-DT and Instructional domain). Results indicate that while our parser does not yet match the performance of a parser trained and tested on the same dataset (intra-domain), it does perform remarkably well on the much more difficult and arguably more useful task of inter-domain discourse structure prediction, where the parser is trained on one domain and tested/applied on another one.

Patrick Huber, Jan Niehues, Alex Waibel

We present a new recurrent neural network topology to enhance state-of-the-art machine learning systems by incorporating a broader context. Our approach overcomes recent limitations with extended narratives through a multi-layered computational approach to generate an abstract context representation. Therefore, the developed system captures the narrative on word-level, sentence-level, and context-level. Through the hierarchical set-up, our proposed model summarizes the most salient information on each level and creates an abstract representation of the extended context. We subsequently use this representation to enhance neural language processing systems on the task of semantic error detection. To show the potential of the newly introduced topology, we compare the approach against a context-agnostic set-up including a standard neural language model and a supervised binary classification network. The performance measures on the error detection task show the advantage of the hierarchical context-aware topologies, improving the baseline by 12.75% relative for unsupervised models and 20.37% relative for supervised models.

Patrick Huber, Jan Niehues, Alex Waibel

We present a new approach to evaluate computational models for the task of text understanding by the means of out-of-context error detection. Through the novel design of our automated modification process, existing large-scale data sources can be adopted for a vast number of text understanding tasks. The data is thereby altered on a semantic level, allowing models to be tested against a challenging set of modified text passages that require to comprise a broader narrative discourse. Our newly introduced task targets actual real-world problems of transcription and translation systems by inserting authentic out-of-context errors. The automated modification process is applied to the 2016 TEDTalk corpus. Entirely automating the process allows the adoption of complete datasets at low cost, facilitating supervised learning procedures and deeper networks to be trained and tested. To evaluate the quality of the modification algorithm a language model and a supervised binary classification model are trained and tested on the altered dataset. A human baseline evaluation is examined to compare the results with human performance. The outcome of the evaluation task indicates the difficulty to detect semantic errors for machine-learning algorithms and humans, showing that the errors cannot be identified when limited to a single sentence.