Jump to Content

Dara Bahri

Personal website: dara.run
Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, desc
  • Year
  • Year, desc
    Preview abstract Existing pre-trained models are generally geared towards a particular class of problems. To date, there seems to be still no consensus on what the right architecture and pre-training setup should be. This paper presents a unified framework for pre-training models that are universally effective across datasets and setups. We begin by disentangling architectural archetypes with pre-training objectives – two concepts that are commonly conflated. Next, we present a generalized and unified perspective for self-supervision in NLP and show how different pre-training objectives can be cast as one another and how interpolating between different objectives can be effective. We then propose Mixture-of-Denoisers (MoD), a pretraining objective that combines diverse pre-training paradigms together. We furthermore introduce a notion of mode switching, wherein downstream fine-tuning is associated with specific pre-training schemes. We conduct extensive ablative experiments to compare multiple pre-training objectives and find that our method pushes the Pareto-frontier by outperforming T5 and/or GPT-like models across multiple diverse setups. Finally, by scaling our model up to 20B parameters, we achieve SOTA performance on 50 well-established supervised NLP tasks ranging from language generation (with automated and human evaluation), language understanding, text classification, question answering, commonsense reasoning, long text reasoning, structured knowledge grounding and information retrieval. Our model also achieve strong results at in-context learning, outperforming 175B GPT-3 on zero-shot SuperGLUE and tripling the performance of T5-XXL on oneshot summarization. Finally, we show that UL2 20B works well with chain-ofthought prompting and reasoning tasks, making it an appealing choice for research into reasoning at a small to medium scale of 20B parameters. We publicly release Flax-based T5X model checkpoints for the 20B model. View details
    Preview abstract In this paper, we demonstrate that information retrieval can be accomplished with a single Transformer, in which all information about the corpus is encoded in the parameters of the model. To this end, we introduce the Differentiable Search Index (DSI), a new paradigm that learns a text-to-text model that maps string queries directly to relevant docids; in other words, a DSI model answers queries directly using only its parameters, dramatically simplifying the whole retrieval process. We study variations in how documents and their identifiers are represented, variations in training procedures, and the interplay between models and corpus sizes. Experiments demonstrate that given appropriate design choices, DSI significantly outperforms strong baselines such as dual encoder models. Moreover, DSI demonstrates strong generalization capabilities, outperforming a BM25 baseline in a zero-shot setup. View details
    Preview abstract State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a block scoring network. We additionally introduce Charformer, a deep Transformer model that integrates GBST and operates on the byte level. Via extensive experiments on English GLUE, multilingual, and noisy text datasets, we show that Charformer outperforms a series of competitive byte-level baselines while generally performing on par and sometimes outperforming subword-based models. Additionally, Charformer is fast, improving the speed of both vanilla byte-level and subword-level Transformers by 28%-100% while maintaining competitive quality. We believe this work paves the way for highly performant token-free models that are trained completely end-to-end. View details
    Preview abstract Despite the recent success of multi-task learning and transfer learning for natural language processing (NLP), few works have systematically studied the effect of scaling up the number of tasks during pre-training. Towards this goal, this paper introduces ExMix (Extreme Mixture): a massive collection of 107 supervised NLP tasks across diverse domains and task-families. Using ExMix, we study the effect of multi-task pre-training at the largest scale to date, and analyze co-training transfer amongst common families of tasks. Through this analysis, we show that manually curating an ideal set of tasks for multi-task pre-training is not straightforward, and that multi-task scaling can vastly improve models on its own. Finally, we propose ExT5: a model pre-trained using a multi-task objective of self-supervised span denoising and supervised ExMix. Via extensive experiments, we show that ExT5 outperforms strong T5 baselines on SuperGLUE, GEM, Rainbow, Closed-Book QA tasks, and several tasks outside of ExMix. ExT5 also significantly improves sample efficiency while pre-training. View details
    Preview abstract Recent advances in Transformer-based large language models (LLMs) achieved significant performance improvements across many tasks. These gains come with a drastic increase in the models' size, leading to slow and costly use at inference time. In practice, however, the series of generations made by LLMs is composed of varying levels of difficulty. While certain predictions truly benefit from the models' full capacity, other continuations are more trivial and can be solved with reduced compute. In this work, we introduce Confident Adaptive Language Modeling (CALM), a method for dynamically allocating different amounts of compute per example and per generation timestep. Early exit decoding involves several challenges that we address here, such as: (1) what confidence measure to use; (2) connecting sequence-level constraints to local per-token exit decisions; and (3) attending back to missing hidden representations due to early exits in previous tokens. Through theoretical analysis and empirical experiments on three diverse generation tasks, we demonstrate the efficacy of our method in reliably reducing compute while maintaining high performance. View details
    Preview abstract State-of-the-art neural models typically encode document-query pairs using cross-attention for re-ranking. To this end, models generally utilize an encoder-only (like BERT) paradigm or an encoder-decoder (like T5) approach. These paradigms, however, are not without flaws, i.e., running the model on all query-document pairs at inference-time incurs a significant computational cost. This paper proposes a new training and inference paradigm for re-ranking. We propose to finetune a pretrained encoder-decoder model using in the form of document to query generation. Subsequently, we show that this encoder-decoder architecture can be decomposed into a decoder-only language model during inference. This results in significant inference time speedups since the decoder-only architecture only needs to learn to interpret static encoder embeddings during inference. Our experiments show that this new paradigm achieves results that are comparable to the more expensive cross-attention ranking approaches while being up to 6.8X faster. We believe this work paves the way for more efficient neural rankers that leverage large pretrained models. View details
    Preview abstract In real-world systems, models are frequently updated as more data becomes available, and in addition to achieving high accuracy, the goal is to also maintain a low difference in predictions compared to the base model (i.e. predictive churn). If model retraining results in vastly different behavior, then it could cause negative effects in downstream systems, especially if this churn can be avoided with limited impact on model accuracy. In this paper, we show an equivalence between training with distillation using the base model as the teacher and training with an explicit constraint on the predictive churn. We then show that distillation performs strongly for low churn training against a number of recent baselines on a wide range of datasets and model architectures, including fully-connected networks, convolutional networks, and transformers. View details
    Preview abstract Self-supervised contrastive representation learning has proved incredibly successful in the vision and natural language domains, enabling state-of-the-art performance with orders of magnitude less labeled data. However, such methods are domain-specific and little has been done to leverage this technique on real-world tabular datasets. We propose SCARF, a simple, widely-applicable technique for contrastive learning, where views are formed by corrupting a random subset of features. When applied to pre-train deep neural networks on the 69 real-world, tabular classification datasets from the OpenML-CC18 benchmark, SCARF not only improves classification accuracy in the fully-supervised setting but does so also in the presence of label noise and in the semi-supervised setting where only a fraction of the available training data is labeled. We show that SCARF complements existing strategies and outperforms alternatives like autoencoders. We conduct comprehensive ablations, detailing the importance of a range of factors. View details
    Sharpness-Aware Minimization Improves Language Model Generalization
    Yi Tay
    Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (2022), pp. 7360-7371
    Preview abstract The allure of superhuman-level capabilities has led to considerable interest in language models like GPT-3 and T5, wherein the research has, by and large, revolved around new model architectures, training tasks, and loss objectives, along with substantial engineering efforts to scale up model capacity and dataset size. Comparatively little work has been done to improve the generalization of these models through better optimization. In this work, we show that Sharpness-Aware Minimization (SAM), a recently proposed optimization procedure that encourages convergence to flatter minima, can substantially improve the generalization of language models without much computational overhead. We show that SAM is able to boost performance on SuperGLUE, GLUE, Web Questions, Natural Questions, Trivia QA, and TyDiQA, with particularly large gains when training data for these tasks is limited. View details
    Preview abstract Large generative language models such as GPT-2 are well-known for not only their ability to generate highly realistic text but also in their utility for common downstream tasks. However, how and in what settings one can best leverage these powerful language models is still a nascent research question. In this work, we explore their use in predicting ``language quality'', a notion of coherence and understandability of text. Our key finding is that, when trained in a self-discriminating fashion, large language models emerge as unsupervised predictors for such language quality. This enables fast bootstrapping of quality indicators in a low-resource setting. We conduct extensive qualitative and quantitative analysis over 500 million web articles, the largest-scale study conducted on this topic. View details
    Preview abstract When experiencing an information need, users want to engage with a domain expert, but often turn to an information retrieval system, such as a search engine, instead. Classical information retrieval systems do not answer information needs directly, but instead provide references to (hopefully authoritative) answers. Successful question answering systems offer a limited corpus created on-demand by human experts, which is neither timely nor scalable. Pre-trained language models, by contrast, are capable of directly generating prose that may be responsive to an information need, but at present they are dilettantes rather than domain experts -- they do not have a true understanding of the world, they are prone to hallucinating, and crucially they are incapable of justifying their utterances by referring to supporting documents in the corpus they were trained over. This paper examines how ideas from classical information retrieval and pre-trained language models can be synthesized and evolved into systems that truly deliver on the promise of domain expert advice. View details
    Preview abstract Achieving state-of-the-art performance on natural language understanding tasks typically relies on fine-tuning a fresh model for every task. Consequently, this approach leads to a higher overall parameter cost, along with higher technical maintenance for serving multiple models. Learning a single multi-task model that is able to do well for all the tasks has been a challenging and yet attractive proposition. In this paper, we propose \textsc{HyperGrid}, a new approach for highly effective multi-task learning. The proposed approach is based on a decomposable hypernetwork that learns grid-wise projections, which helps to specialize regions in weight matrices for different tasks. In order to construct the proposed hyper projection, our method learns the interactions and composition between a global state and a local task-specific state. We apply our proposed \textsc{HyperGrid} on the current state-of-the-art T5 model, yielding optimistic and strong gains across GLUE and SuperGLUE benchmarks when trained in a single model multi-tasking setup. Our method helps to bridge the gap between the single-task finetune methods and the single model multi-tasking approaches View details
    Preview abstract In the era of pretrained language models, transformers are the defacto choice of model architectures. While recent works have shown promise in entirely convolutional based architectures, these CNN-based models have not been widely adopted or evaluated under the pretrain-finetune paradigm. In the context of language models, are convolutional models competitive when pretrained? This paper investigates this research question and presents several interesting findings. Across a set of extensive experiments, our findings show that CNN-based pretrained models are highly competitive and outperform Transformer-based pretrained models in certain scenarios, albeit with caveats. Overall, the findings of this paper should implore the broader academic community to perhaps not conflate pretraining advances with architectural advances and both set of techniques could be studied in isolation. View details
    Preview abstract The grammar of the natural language has two major classes: Dependency grammar that models are one-to-one correspondences between words, and Constituency grammar that models the assembly of one or several corresponded words. While previous unsupervised parsing methods mostly focus on inducing one class of grammars, we introduce a novel model, StructFormer, that can induce dependency and constituency structure at the same time. In order to achieve this, we propose a new self-attention mechanism with novel hierarchical and dependency constraints. Experiment results show that our model can achieve strong results on Unsupervised Constituency parsing, Unsupervised Dependency Parsing and Masked Language Modeling. View details
    Preview abstract Transformers do not scale very well to long sequence lengths largely because of quadratic self-attention complexity. In the recent months, a wide spectrum of efficient, fast Transformers have been proposed to tackle this problem, more often than not claiming superior or comparable performance to vanilla Transformer models. To this date, there is no well-established consensus on how to evaluate this class of models. Moreover, inconsistent benchmarking on a wide spectrum of tasks and datasets makes it difficult to assess relative performance amongst many models. This paper proposes a systematic and unified benchmark, LRA a benchmark specifically focused on evaluating model quality under long-context scenarios. Our benchmark is a suite of tasks consisting of sequences ranging from $1K$ to $16K$ tokens, encompassing a wide range of data types and modalities such as text, natural and synthetic images, and mathematical expressions requiring similarity, structural and visual-spatial reasoning. We systematically evaluate ten well established long-range Transformer models (Reformers, Linformers, Linear Transformers, Sinkhorn Transformers, Performers, Synthesizers, Sparse Transformers and Longformers) on our newly proposed benchmark suite. LRA paves the way towards better understanding this class of efficient Transformer models, facilitates more research in this direction, and presents new challenging tasks to tackle. View details
    Preview abstract This paper proposes Omnidirectional Representations from Transformers (\textsc{OmniNet}). In OmniNet, instead of maintaing a strictly horizontal receptive field, each token is allowed to attend to all tokens in the entire network. This process can also be interpreted as a form of extreme or intensive attention mechanism that has the receptive field of the entire width and depth of the network. To this end, the omnidirection attention is learned via a meta-learner, which is essentially another self-attention based model. In order to mitigate the computationally expensive costs of full receptive field attention, we leverage efficient self-attention models such as kernel-based \cite{choromanski2020rethinking}, low-rank attention \cite{wang2020linformer} and/or Big Bird \cite{zaheer2020big} as the meta-learner. We conduct extensive experiments on autoregressive language modeling (LM1B, C4), Machine Translation, Long Range Arena (LRA) and Image Recognition, showing that OmniNet not only achieves considerable improvements when equipped with both sequence-based (1D) Transformers but also on image recognition (finetuning and few shot learning) tasks. OmniNet also achieves state-of-the-art performance on LM1B, WMT'14 En-De/En-Fr and Long Range Arena. View details
    Preview abstract We propose Sparse Sinkhorn Attention, a new efficient and sparse method for learning to attend. Our method is based on differentiable sorting of internal representations. Concretely, we introduce a meta sorting network that learns to generate latent permutations over sequences. Given sorted sequences, we are then able to compute quasi-global attention with only local windows, improving the memory efficiency of the attention module. To this end, we propose new algorithmic innovations such as Causal Sinkhorn Balancing and SortCut, a dynamic sequence truncation method for tailoring Sinkhorn Attention for encoding and/or decoding purposes. Via extensive experiments on algorithmic seq2seq sorting, language modeling, pixel-wise image generation, document classification and natural language inference, we demonstrate that our Sinkhorn Attention remains competitive to the vanilla attention, consistently outperforming recently proposed efficient Transformer models such as Sparse Transformers, while retaining memory efficiency. View details
    Preview abstract This paper seeks to develop a deeper understanding of the fundamental properties of neural text generations models. Concretely, the study of artifacts that emerge in machine generated text as a result of modeling choices is a nascent research area. To this end, the extent and degree to which these artifacts surface in generated text is still unclear. In the spirit of better understanding generative text models and their artifacts, we propose the new task of distinguishing which of several variants of a given model generated some piece of text. Specifically, we conduct an extensive suite of diagnostic tests to observe whether modeling choices (e.g., sampling methods, top-$k$ probabilities, model architectures, etc.) leave detectable artifacts in the text they generate. Our key finding, which is backed by a rigorous set of experiments, is that such artifacts are present and that different modeling choices can be inferred by looking at generated text alone. This suggests that neural text generators may actually be more sensitive to various modeling choices than previously thought. View details
    Preview abstract Work in information retrieval has traditionally been focused on ranking and relevance: for a user's query, fetch some number of results, ordered by relevance to the user. However, the problem of determining how many results to return, i.e. how to optimally truncate the ranked result list, has received far less attention despite being of critical importance in a range of applications. Such truncation is a balancing act between the overall relevance, or usefulness, of the results with the user cost of processing more results. In this work, we propose Choppy, an assumption-free model based on the widely successful Transformer architecture in NLP, to the ranked-list truncation problem. Needing nothing more than the relevance scores of the results, the model uses a powerful multi-head attention mechanism to directly optimize any user-defined target IR metric. We show Choppy improves upon recent, state-of-the-art baselines on Robust04. View details
    Preview abstract We investigate machine learning models that can provide diminishing returns and accelerating returns guarantees to capture prior knowledge or policies about how outputs should depend on inputs. We show that one can build flexible, nonlinear, multi-dimensional models using lattice functions with any combination of concavity/convexity and monotonicity constraints on any subsets of features, and compare to new shape-constrained neural networks. We demonstrate on real-world examples that these shape constrained models can provide tuning-free regularization and improve model understandability. View details
    No Results Found