Priyanka Agrawal

Joshua Maynez, Priyanka Agrawal, Sebastian Gehrmann · deepmind

Pre-trained large language models (PLMs) underlie most new developments in natural language processing. They have shifted the field from application-specific model pipelines to a single model that is adapted to a wide range of tasks. Autoregressive PLMs like GPT-3 or PaLM, alongside techniques like few-shot learning, have additionally shifted the output modality to generation instead of classification or regression. Despite their ubiquitous use, the generation quality of language models is rarely evaluated when these models are introduced. Additionally, it is unclear how existing generation tasks--while they can be used to compare systems at a high level--relate to the real world use cases for which people have been adopting them. In this work, we discuss how to adapt existing application-specific generation benchmarks to PLMs and provide an in-depth, empirical study of the limitations and capabilities of PLMs in natural language generation tasks along dimensions such as scale, architecture, input and output language. Our results show that PLMs differ in their applicability to different data regimes and their generalization to multiple languages and inform which PLMs to use for a given generation task setup. We share best practices to be taken into consideration when benchmarking generation capabilities during the development of upcoming PLMs.

Priyanka Agrawal, Chris Alberti, Fantine Huot et al.

The availability of large, high-quality datasets has been one of the main drivers of recent progress in question answering (QA). Such annotated datasets however are difficult and costly to collect, and rarely exist in languages other than English, rendering QA technology inaccessible to underrepresented languages. An alternative to building large monolingual training datasets is to leverage pre-trained language models (PLMs) under a few-shot learning setting. Our approach, QAmeleon, uses a PLM to automatically generate multilingual data upon which QA models are trained, thus avoiding costly annotation. Prompt tuning the PLM for data synthesis with only five examples per language delivers accuracy superior to translation-based baselines, bridges nearly 60% of the gap between an English-only baseline and a fully supervised upper bound trained on almost 50,000 hand labeled examples, and always leads to substantial improvements compared to fine-tuning a QA model directly on labeled examples in low resource settings. Experiments on the TyDiQA-GoldP and MLQA benchmarks show that few-shot prompt tuning for data synthesis scales across languages and is a viable alternative to large-scale annotation.

Alexandra Chronopoulou, Jonas Pfeiffer, Joshua Maynez et al.

Parameter-efficient fine-tuning (PEFT) using labeled task data can significantly improve the performance of large language models (LLMs) on the downstream task. However, there are 7000 languages in the world and many of these languages lack labeled data for real-world language generation tasks. In this paper, we propose to improve zero-shot cross-lingual transfer by composing language or task specialized parameters. Our method composes language and task PEFT modules via element-wise arithmetic operations to leverage unlabeled data and English labeled data. We extend our approach to cases where labeled data from more languages is available and propose to arithmetically compose PEFT modules trained on languages related to the target. Empirical results on summarization demonstrate that our method is an effective strategy that obtains consistent gains using minimal training of PEFT modules.

Aileen Cheng, Alon Jacovi, Amir Globerson et al.

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models' world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model's overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts .

Gemini Team, Petko Georgiev, Ving Ian Lei et al. · deepmind, mila

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Chaitanya Malaviya, Priyanka Agrawal, Kuzman Ganchev et al.

Experts in various fields routinely perform methodical writing tasks to plan, organize, and report their work. From a clinician writing a differential diagnosis for a patient, to a teacher writing a lesson plan for students, these tasks are pervasive, requiring to methodically generate structured long-form output for a given input. We develop a typology of methodical tasks structured in the form of a task objective, procedure, input, and output, and introduce DoLoMiTes, a novel benchmark with specifications for 519 such tasks elicited from hundreds of experts from across 25 fields. Our benchmark further contains specific instantiations of methodical tasks with concrete input and output examples (1,857 in total) which we obtain by collecting expert revisions of up to 10 model-generated examples of each task. We use these examples to evaluate contemporary language models highlighting that automating methodical tasks is a challenging long-form generation problem, as it requires performing complex inferences, while drawing upon the given context as well as domain knowledge.

Yukti Makhija, Priyanka Agrawal, Rishi Saket et al.

Large language models (LLMs) are being increasingly tuned to power complex generation tasks such as writing, fact-seeking, querying and reasoning. Traditionally, human or model feedback for evaluating and further tuning LLM performance has been provided at the response level, enabling faster and more cost-effective assessments. However, recent works (Amplayo et al. [2022], Wu et al. [2023]) indicate that sentence-level labels may provide more accurate and interpretable feedback for LLM optimization. In this work, we introduce methods to disaggregate response-level labels into sentence-level (pseudo-)labels. Our approach leverages multiple instance learning (MIL) and learning from label proportions (LLP) techniques in conjunction with prior information (e.g., document-sentence cosine similarity) to train a specialized model for sentence-level scoring. We also employ techniques which use model predictions to pseudo-label the train-set at the sentence-level for model training to further improve performance. We conduct extensive evaluations of our methods across six datasets and four tasks: retrieval, question answering, summarization, and math reasoning. Our results demonstrate improved performance compared to multiple baselines across most of these tasks. Our work is the first to develop response-level feedback to sentence-level scoring techniques, leveraging sentence-level prior information, along with comprehensive evaluations on multiple tasks as well as end-to-end finetuning evaluation showing performance comparable to a model trained on fine-grained human annotated labels.

Gemini Team, Rohan Anil, Sebastian Borgeaud et al.

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

Fantine Huot, Joshua Maynez, Chris Alberti et al.

Cross-lingual summarization consists of generating a summary in one language given an input document in a different language, allowing for the dissemination of relevant content across speakers of other languages. The task is challenging mainly due to the paucity of cross-lingual datasets and the compounded difficulty of summarizing and translating. This work presents $μ$PLAN, an approach to cross-lingual summarization that uses an intermediate planning step as a cross-lingual bridge. We formulate the plan as a sequence of entities capturing the summary's content and the order in which it should be communicated. Importantly, our plans abstract from surface form: using a multilingual knowledge base, we align entities to their canonical designation across languages and generate the summary conditioned on this cross-lingual bridge and the input. Automatic and human evaluation on the XWikis dataset (across four language pairs) demonstrates that our planning objective achieves state-of-the-art performance in terms of informativeness and faithfulness. Moreover, $μ$PLAN models improve the zero-shot transfer to new cross-lingual language pairs compared to baselines without a planning component.

Emmanouil Stergiadis, Priyanka Agrawal, Oliver Squire

We propose an adaptation of the curriculum training framework, applicable to state-of-the-art meta learning techniques for few-shot classification. Curriculum-based training popularly attempts to mimic human learning by progressively increasing the training complexity to enable incremental concept learning. As the meta-learner's goal is learning how to learn from as few samples as possible, the exact number of those samples (i.e. the size of the support set) arises as a natural proxy of a given task's difficulty. We define a simple yet novel curriculum schedule that begins with a larger support size and progressively reduces it throughout training to eventually match the desired shot-size of the test setup. This proposed method boosts the learning efficiency as well as the generalization capability. Our experiments with the MAML algorithm on two few-shot image classification tasks show significant gains with the curriculum training framework. Ablation studies corroborate the independence of our proposed method from the model architecture as well as the meta-learning hyperparameters

Subhojeet Pramanik, Priyanka Agrawal, Aman Hussain

Transformer is a popularly used neural network architecture, especially for language understanding. We introduce an extended and unified architecture that can be used for tasks involving a variety of modalities like image, text, videos, etc. We propose a spatio-temporal cache mechanism that enables learning spatial dimension of the input in addition to the hidden states corresponding to the temporal input sequence. The proposed architecture further enables a single model to support tasks with multiple input modalities as well as asynchronous multi-task learning, thus we refer to it as OmniNet. For example, a single instance of OmniNet can concurrently learn to perform the tasks of part-of-speech tagging, image captioning, visual question answering and video activity recognition. We demonstrate that training these four tasks together results in about three times compressed model while retaining the performance in comparison to training them individually. We also show that using this neural network pre-trained on some modalities assists in learning unseen tasks such as video captioning and video question answering. This illustrates the generalization capacity of the self-attention mechanism on the spatio-temporal cache present in OmniNet.

Priyanka Agrawal, Parag Jain, Ayushi Dalmia et al.

Semantic parsing over multiple knowledge bases enables a parser to exploit structural similarities of programs across the multiple domains. However, the fundamental challenge lies in obtaining high-quality annotations of (utterance, program) pairs across various domains needed for training such models. To overcome this, we propose a novel framework to build a unified multi-domain enabled semantic parser trained only with weak supervision (denotations). Weakly supervised training is particularly arduous as the program search space grows exponentially in a multi-domain setting. To solve this, we incorporate a multi-policy distillation mechanism in which we first train domain-specific semantic parsers (teachers) using weak supervision in the absence of the ground truth programs, followed by training a single unified parser (student) from the domain specific policies obtained from these teachers. The resultant semantic parser is not only compact but also generalizes better, and generates more accurate programs. It further does not require the user to provide a domain label while querying. On the standard Overnight dataset (containing multiple domains), we demonstrate that the proposed model improves performance by 20% in terms of denotation accuracy in comparison to baseline techniques.

Anirban Laha, Saneem A. Chemmengath, Priyanka Agrawal et al.

Converting an n-dimensional vector to a probability distribution over n objects is a commonly used component in many machine learning tasks like multiclass classification, multilabel classification, attention mechanisms etc. For this, several probability mapping functions have been proposed and employed in literature such as softmax, sum-normalization, spherical softmax, and sparsemax, but there is very little understanding in terms how they relate with each other. Further, none of the above formulations offer an explicit control over the degree of sparsity. To address this, we develop a unified framework that encompasses all these formulations as special cases. This framework ensures simple closed-form solutions and existence of sub-gradients suitable for learning via backpropagation. Within this framework, we propose two novel sparse formulations, sparsegen-lin and sparsehourglass, that seek to provide a control over the degree of desired sparsity. We further develop novel convex loss functions that help induce the behavior of aforementioned formulations in the multilabel classification setting, showing improved performance. We also demonstrate empirically that the proposed formulations, when used to compute attention weights, achieve better or comparable performance on standard seq2seq tasks like neural machine translation and abstractive summarization.

Amar Prakash Azad, Dinesh Garg, Priyanka Agrawal et al.

The goal behind Domain Adaptation (DA) is to leverage the labeled examples from a source domain so as to infer an accurate model in a target domain where labels are not available or in scarce at the best. A state-of-the-art approach for the DA is due to (Ganin et al. 2016), known as DANN, where they attempt to induce a common representation of source and target domains via adversarial training. This approach requires a large number of labeled examples from the source domain to be able to infer a good model for the target domain. However, in many situations obtaining labels in the source domain is expensive which results in deteriorated performance of DANN and limits its applicability in such scenarios. In this paper, we propose a novel approach to overcome this limitation. In our work, we first establish that DANN reduces the original DA problem into a semi-supervised learning problem over the space of common representation. Next, we propose a learning approach, namely TransDANN, that amalgamates adversarial learning and transductive learning to mitigate the detrimental impact of limited source labels and yields improved performance. Experimental results (both on text and images) show a significant boost in the performance of TransDANN over DANN under such scenarios. We also provide theoretical justification for the performance boost.

Parag Jain, Priyanka Agrawal, Abhijit Mishra et al.

Existing Natural Language Generation (NLG) systems are weak AI systems and exhibit limited capabilities when language generation tasks demand higher levels of creativity, originality and brevity. Effective solutions or, at least evaluations of modern NLG paradigms for such creative tasks have been elusive, unfortunately. This paper introduces and addresses the task of coherent story generation from independent descriptions, describing a scene or an event. Towards this, we explore along two popular text-generation paradigms -- (1) Statistical Machine Translation (SMT), posing story generation as a translation problem and (2) Deep Learning, posing story generation as a sequence to sequence learning problem. In SMT, we chose two popular methods such as phrase based SMT (PB-SMT) and syntax based SMT (SYNTAX-SMT) to `translate' the incoherent input text into stories. We then implement a deep recurrent neural network (RNN) architecture that encodes sequence of variable length input descriptions to corresponding latent representations and decodes them to produce well formed comprehensive story like summaries. The efficacy of the suggested approaches is demonstrated on a publicly available dataset with the help of popular machine translation and summarization evaluation metrics.

Lavanya Sita Tekumalla, Priyanka Agrawal, Indrajit Bhattacharya

Dirichlet Process(DP) is a Bayesian non-parametric prior for infinite mixture modeling, where the number of mixture components grows with the number of data items. The Hierarchical Dirichlet Process (HDP), is an extension of DP for grouped data, often used for non-parametric topic modeling, where each group is a mixture over shared mixture densities. The Nested Dirichlet Process (nDP), on the other hand, is an extension of the DP for learning group level distributions from data, simultaneously clustering the groups. It allows group level distributions to be shared across groups in a non-parametric setting, leading to a non-parametric mixture of mixtures. The nCRF extends the nDP for multilevel non-parametric mixture modeling, enabling modeling topic hierarchies. However, the nDP and nCRF do not allow sharing of distributions as required in many applications, motivating the need for multi-level non-parametric admixture modeling. We address this gap by proposing multi-level nested HDPs (nHDP) where the base distribution of the HDP is itself a HDP at each level thereby leading to admixtures of admixtures at each level. Because of couplings between various HDP levels, scaling up is naturally a challenge during inference. We propose a multi-level nested Chinese Restaurant Franchise (nCRF) representation for the nested HDP, with which we outline an inference algorithm based on Gibbs Sampling. We evaluate our model with the two level nHDP for non-parametric entity topic modeling where an inner HDP creates a countably infinite set of topic mixtures and associates them with author entities, while an outer HDP associates documents with these author entities. In our experiments on two real world research corpora, the nHDP is able to generalize significantly better than existing models and detect missing author entities with a reasonable level of accuracy.