Yizhong Wang

Yizhong Wang, Yeganeh Kordi, Swaroop Mishra et al. · allen-ai, uw

Large "instruction-tuned" language models (i.e., finetuned to respond to instructions) have demonstrated a remarkable ability to generalize zero-shot to new tasks. Nevertheless, they depend heavily on human-written instruction data that is often limited in quantity, diversity, and creativity, therefore hindering the generality of the tuned model. We introduce Self-Instruct, a framework for improving the instruction-following capabilities of pretrained language models by bootstrapping off their own generations. Our pipeline generates instructions, input, and output samples from a language model, then filters invalid or similar ones before using them to finetune the original model. Applying our method to the vanilla GPT3, we demonstrate a 33% absolute improvement over the original model on Super-NaturalInstructions, on par with the performance of InstructGPT-001, which was trained with private user data and human annotations. For further evaluation, we curate a set of expert-written instructions for novel tasks, and show through human evaluation that tuning GPT3 with Self-Instruct outperforms using existing public instruction datasets by a large margin, leaving only a 5% absolute gap behind InstructGPT-001. Self-Instruct provides an almost annotation-free method for aligning pre-trained language models with instructions, and we release our large synthetic dataset to facilitate future studies on instruction tuning. Our code and data are available at https://github.com/yizhongw/self-instruct.

Joel Jang, Seungone Kim, Bill Yuchen Lin et al. · allen-ai, cmu

While Reinforcement Learning from Human Feedback (RLHF) aligns Large Language Models (LLMs) with general, aggregate human preferences, it is suboptimal for learning diverse, individual perspectives. In this work, we study Reinforcement Learning from Personalized Human Feedback (RLPHF) problem, wherein LLMs are aligned to multiple (sometimes conflicting) preferences by modeling alignment as a Multi-Objective Reinforcement Learning (MORL) problem. Compared to strong single-objective baselines, we show that we can achieve personalized alignment by decomposing preferences into multiple dimensions. These dimensions are defined based on personalizations that are declared as desirable by the user. In this work, we show that they can be efficiently trained independently in a distributed manner and combined effectively post-hoc through parameter merging. The code is available at https://github.com/joeljang/RLPHF.

Yizhong Wang, Hamish Ivison, Pradeep Dasigi et al. · allen-ai, uw

In this work we explore recent advances in instruction-tuning language models on a range of open instruction-following datasets. Despite recent claims that open models can be on par with state-of-the-art proprietary models, these claims are often accompanied by limited evaluation, making it difficult to compare models across the board and determine the utility of various resources. We provide a large set of instruction-tuned models from 6.7B to 65B parameters in size, trained on 12 instruction datasets ranging from manually curated (e.g., OpenAssistant) to synthetic and distilled (e.g., Alpaca) and systematically evaluate them on their factual knowledge, reasoning, multilinguality, coding, and open-ended instruction following abilities through a collection of automatic, model-based, and human-based metrics. We further introduce Tülu, our best performing instruction-tuned model suite finetuned on a combination of high-quality open resources. Our experiments show that different instruction-tuning datasets can uncover or enhance specific skills, while no single dataset (or combination) provides the best performance across all evaluations. Interestingly, we find that model and human preference-based evaluations fail to reflect differences in model capabilities exposed by benchmark-based evaluations, suggesting the need for the type of systemic evaluation performed in this work. Our evaluations show that the best model in any given evaluation reaches on average 87% of ChatGPT performance, and 73% of GPT-4 performance, suggesting that further investment in building better base models and instruction-tuning data is required to close the gap. We release our instruction-tuned models, including a fully finetuned 65B Tülu, along with our code, data, and evaluation framework at https://github.com/allenai/open-instruct to facilitate future research.

Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi et al. · allen-ai, amazon-science

How well can NLP models generalize to a variety of unseen tasks when provided with task instructions? To address this question, we first introduce Super-NaturalInstructions, a benchmark of 1,616 diverse NLP tasks and their expert-written instructions. Our collection covers 76 distinct task types, including but not limited to classification, extraction, infilling, sequence tagging, text rewriting, and text composition. This large and diverse collection of tasks enables rigorous benchmarking of cross-task generalization under instructions -- training models to follow instructions on a subset of tasks and evaluating them on the remaining unseen ones. Furthermore, we build Tk-Instruct, a transformer model trained to follow a variety of in-context instructions (plain language task definitions or k-shot examples). Our experiments show that Tk-Instruct outperforms existing instruction-following models such as InstructGPT by over 9% on our benchmark despite being an order of magnitude smaller. We further analyze generalization as a function of various scaling parameters, such as the number of observed tasks, the number of instances per task, and model sizes. We hope our dataset and model facilitate future progress towards more general-purpose NLP models.

Akari Asai, Zeqiu Wu, Yizhong Wang et al. · uw

Despite their remarkable capabilities, large language models (LLMs) often produce responses containing factual inaccuracies due to their sole reliance on the parametric knowledge they encapsulate. Retrieval-Augmented Generation (RAG), an ad hoc approach that augments LMs with retrieval of relevant knowledge, decreases such issues. However, indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation. We introduce a new framework called Self-Reflective Retrieval-Augmented Generation (Self-RAG) that enhances an LM's quality and factuality through retrieval and self-reflection. Our framework trains a single arbitrary LM that adaptively retrieves passages on-demand, and generates and reflects on retrieved passages and its own generations using special tokens, called reflection tokens. Generating reflection tokens makes the LM controllable during the inference phase, enabling it to tailor its behavior to diverse task requirements. Experiments show that Self-RAG (7B and 13B parameters) significantly outperforms state-of-the-art LLMs and retrieval-augmented models on a diverse set of tasks. Specifically, Self-RAG outperforms ChatGPT and retrieval-augmented Llama2-chat on Open-domain QA, reasoning and fact verification tasks, and it shows significant gains in improving factuality and citation accuracy for long-form generations relative to these models.

Hongjin Su, Weijia Shi, Jungo Kasai et al. · allen-ai, uw

We introduce INSTRUCTOR, a new method for computing text embeddings given task instructions: every text input is embedded together with instructions explaining the use case (e.g., task and domain descriptions). Unlike encoders from prior work that are more specialized, INSTRUCTOR is a single embedder that can generate text embeddings tailored to different downstream tasks and domains, without any further training. We first annotate instructions for 330 diverse tasks and train INSTRUCTOR on this multitask mixture with a contrastive loss. We evaluate INSTRUCTOR on 70 embedding evaluation tasks (66 of which are unseen during training), ranging from classification and information retrieval to semantic textual similarity and text generation evaluation. INSTRUCTOR, while having an order of magnitude fewer parameters than the previous best model, achieves state-of-the-art performance, with an average improvement of 3.4% compared to the previous best results on the 70 diverse datasets. Our analysis suggests that INSTRUCTOR is robust to changes in instructions, and that instruction finetuning mitigates the challenge of training a single model on diverse datasets. Our model, code, and data are available at https://instructor-embedding.github.io.

Yushi Hu, Benlin Liu, Jungo Kasai et al. · allen-ai, uw

Despite thousands of researchers, engineers, and artists actively working on improving text-to-image generation models, systems often fail to produce images that accurately align with the text inputs. We introduce TIFA (Text-to-Image Faithfulness evaluation with question Answering), an automatic evaluation metric that measures the faithfulness of a generated image to its text input via visual question answering (VQA). Specifically, given a text input, we automatically generate several question-answer pairs using a language model. We calculate image faithfulness by checking whether existing VQA models can answer these questions using the generated image. TIFA is a reference-free metric that allows for fine-grained and interpretable evaluations of generated images. TIFA also has better correlations with human judgments than existing metrics. Based on this approach, we introduce TIFA v1.0, a benchmark consisting of 4K diverse text inputs and 25K questions across 12 categories (object, counting, etc.). We present a comprehensive evaluation of existing text-to-image models using TIFA v1.0 and highlight the limitations and challenges of current models. For instance, we find that current text-to-image models, despite doing well on color and material, still struggle in counting, spatial relations, and composing multiple objects. We hope our benchmark will help carefully measure the research progress in text-to-image synthesis and provide valuable insights for further research.

Hamish Ivison, Akshita Bhagia, Yizhong Wang et al. · allen-ai, uw

Recent NLP models have shown the remarkable ability to effectively generalise `zero-shot' to new tasks using only natural language instructions as guidance. However, many of these approaches suffer from high computational costs due to their reliance on concatenating lengthy instructions with every input example, resulting in costly reprocessing of the instruction. To avoid this, we introduce Hypernetworks for INstruction Tuning (HINT), which convert task instructions and examples into parameter-efficient modules inserted into an underlying model using a pretrained text encoder, eliminating the need to include instructions in the model input. The hypernetwork in HINT also produces an encoded instruction, which we concatenate with encoded inputs during decoding to further improve performance. HINT models outperform strong state-of-the-art baselines by over 10% when controlling for compute (measured in FLOPs). By converting instructions into modules, HINT models can effectively disregard the length of instructions and few-shot example inputs in terms of compute usage. As a result, HINT can enhance its performance by up to 25% by incorporating additional few-shot data, while utilizing only up to 5% more compute. This combines the strengths of parameter-efficient fine-tuning and in-context learning.

Hamish Ivison, Yizhong Wang, Valentina Pyatkin et al. · allen-ai, uw

Since the release of TÜLU [Wang et al., 2023b], open resources for instruction tuning have developed quickly, from better base models to new finetuning techniques. We test and incorporate a number of these advances into TÜLU, resulting in TÜLU 2, a suite of improved TÜLU models for advancing the understanding and best practices of adapting pretrained language models to downstream tasks and user preferences. Concretely, we release: (1) TÜLU-V2-mix, an improved collection of high-quality instruction datasets; (2) TÜLU 2, LLAMA-2 models finetuned on the V2 mixture; (3) TÜLU 2+DPO, TÜLU 2 models trained with direct preference optimization (DPO), including the largest DPO-trained model to date (TÜLU 2+DPO 70B); (4) CODE TÜLU 2, CODE LLAMA models finetuned on our V2 mix that outperform CODE LLAMA and its instruction-tuned variant, CODE LLAMA-Instruct. Our evaluation from multiple perspectives shows that the TÜLU 2 suite achieves state-of-the-art performance among open models and matches or exceeds the performance of GPT-3.5-turbo-0301 on several benchmarks. We release all the checkpoints, data, training and evaluation code to facilitate future open efforts on adapting large language models.

Qingqing Cao, Sewon Min, Yizhong Wang et al. · uw

Retrieval augmentation addresses many critical problems in large language models such as hallucination, staleness, and privacy leaks. However, running retrieval-augmented language models (LMs) is slow and difficult to scale due to processing large amounts of retrieved text. We introduce binary token representations (BTR), which use 1-bit vectors to precompute every token in passages, significantly reducing computation during inference. Despite the potential loss of accuracy, our new calibration techniques and training objectives restore performance. Combined with offline and runtime compression, this only requires 127GB of disk space for encoding 3 billion tokens in Wikipedia. Our experiments show that on five knowledge-intensive NLP tasks, BTR accelerates state-of-the-art inference by up to 4x and reduces storage by over 100x while maintaining over 95% task performance.

Kevin Xu, Yeganeh Kordi, Tanay Nayak et al. · uw

Can advanced multi-modal models effectively tackle complex web-based tasks? Such tasks are often found on crowdsourcing platforms, where crowdworkers engage in challenging micro-tasks within web-based environments. Building on this idea, we present TurkingBench, a benchmark consisting of tasks presented as web pages with textual instructions and multi-modal contexts. Unlike previous approaches that rely on artificially synthesized web pages, our benchmark uses natural HTML pages originally designed for crowdsourcing workers to perform various annotation tasks. Each task's HTML instructions are instantiated with different values derived from crowdsourcing tasks, creating diverse instances. This benchmark includes 32.2K instances spread across 158 tasks. To support the evaluation of TurkingBench, we have developed a framework that links chatbot responses to actions on web pages (e.g., modifying a text box, selecting a radio button). We assess the performance of cutting-edge private and open-source models, including language-only and vision-language models (such as GPT4 and InternVL), on this benchmark. Our results show that while these models outperform random chance, there is still significant room for improvement. We hope that this benchmark will drive progress in the evaluation and development of web-based agents.

Keyang Xuan, Peiyang Song, Pan Lu et al.

AI evaluation is undergoing a structural change. Large language models (LLMs) are increasingly deployed as systems that act over time through tools, environments, users, and other agents, while many evaluation practices still inherit assumptions from response-centered benchmarks (e.g., fixed inputs, isolated outputs, and outcome judgments that can be made from a single response). The field has begun to build interactive benchmarks, but the resulting landscape is fragmented: benchmarks differ in what interaction artifacts they admit, how trajectories are scored, and what claims their results support. This position paper argues that interactive evaluation should be treated as a principled evaluation paradigm, not merely a new family of agent benchmarks. Simply adopting previous evaluation paradigms does not suffice. We define evaluation as an autonomous mapping from evidence to judgments, and show that interactive evaluation changes both sides of this mapping: the evidence becomes interaction-generated trajectories, while the evaluation procedure must assess process, recoverability, coordination, robustness, and system-level performance. Building on this definition, we propose a two-axis taxonomy, derive design principles and reporting standards, examine representative scenarios, and analyze how longstanding evaluation challenges reappear at the trajectory level.

Dirk Groeneveld, Iz Beltagy, Pete Walsh et al. · allen-ai, cmu

Language models (LMs) have become ubiquitous in both NLP research and in commercial product offerings. As their commercial importance has surged, the most powerful models have become closed off, gated behind proprietary interfaces, with important details of their training data, architectures, and development undisclosed. Given the importance of these details in scientifically studying these models, including their biases and potential risks, we believe it is essential for the research community to have access to powerful, truly open LMs. To this end, we have built OLMo, a competitive, truly Open Language Model, to enable the scientific study of language models. Unlike most prior efforts that have only released model weights and inference code, we release OLMo alongside open training data and training and evaluation code. We hope this release will empower the open research community and inspire a new wave of innovation.

Nathan Lambert, Jacob Morrison, Valentina Pyatkin et al. · allen-ai, cambridge

Language model post-training is applied to refine behaviors and unlock new skills across a wide range of recent language models, but open recipes for applying these techniques lag behind proprietary ones. The underlying training data and recipes for post-training are simultaneously the most important pieces of the puzzle and the portion with the least transparency. To bridge this gap, we introduce Tulu 3, a family of fully-open state-of-the-art post-trained models, alongside its data, code, and training recipes, serving as a comprehensive guide for modern post-training techniques. Tulu 3, which builds on Llama 3.1 base models, achieves results surpassing the instruct versions of Llama 3.1, Qwen 2.5, Mistral, and even closed models such as GPT-4o-mini and Claude 3.5-Haiku. The training algorithms for our models include supervised finetuning (SFT), Direct Preference Optimization (DPO), and a novel method we call Reinforcement Learning with Verifiable Rewards (RLVR). With Tulu 3, we introduce a multi-task evaluation scheme for post-training recipes with development and unseen evaluations, standard benchmark implementations, and substantial decontamination of existing open datasets on said benchmarks. We conclude with analysis and discussion of training methods that did not reliably improve performance. In addition to the Tulu 3 model weights and demo, we release the complete recipe -- including datasets for diverse core skills, a robust toolkit for data curation and evaluation, the training code and infrastructure, and, most importantly, a detailed report for reproducing and further adapting the Tulu 3 approach to more domains.

Abhika Mishra, Akari Asai, Vidhisha Balachandran et al. · cmu, uw

Large language models (LMs) are prone to generate factual errors, which are often called hallucinations. In this paper, we introduce a comprehensive taxonomy of hallucinations and argue that hallucinations manifest in diverse forms, each requiring varying degrees of careful assessments to verify factuality. We propose a novel task of automatic fine-grained hallucination detection and construct a new evaluation benchmark, FavaBench, that includes about one thousand fine-grained human judgments on three LM outputs across various domains. Our analysis reveals that ChatGPT and Llama2-Chat (70B, 7B) exhibit diverse types of hallucinations in the majority of their outputs in information-seeking scenarios. We train FAVA, a retrieval-augmented LM by carefully creating synthetic data to detect and correct fine-grained hallucinations. On our benchmark, our automatic and human evaluations show that FAVA significantly outperforms ChatGPT and GPT-4 on fine-grained hallucination detection, and edits suggested by FAVA improve the factuality of LM-generated text.

Wenhao Wu, Yizhong Wang, Guangxuan Xiao et al. · uw

Despite the recent progress in long-context language models, it remains elusive how transformer-based models exhibit the capability to retrieve relevant information from arbitrary locations within the long context. This paper aims to address this question. Our systematic investigation across a wide spectrum of models reveals that a special type of attention heads are largely responsible for retrieving information, which we dub retrieval heads. We identify intriguing properties of retrieval heads:(1) universal: all the explored models with long-context capability have a set of retrieval heads; (2) sparse: only a small portion (less than 5\%) of the attention heads are retrieval. (3) intrinsic: retrieval heads already exist in models pretrained with short context. When extending the context length by continual pretraining, it is still the same set of heads that perform information retrieval. (4) dynamically activated: take Llama-2 7B for example, 12 retrieval heads always attend to the required information no matter how the context is changed. The rest of the retrieval heads are activated in different contexts. (5) causal: completely pruning retrieval heads leads to failure in retrieving relevant information and results in hallucination, while pruning random non-retrieval heads does not affect the model's retrieval ability. We further show that retrieval heads strongly influence chain-of-thought (CoT) reasoning, where the model needs to frequently refer back the question and previously-generated context. Conversely, tasks where the model directly generates the answer using its intrinsic knowledge are less impacted by masking out retrieval heads. These observations collectively explain which internal part of the model seeks information from the input tokens. We believe our insights will foster future research on reducing hallucination, improving reasoning, and compressing the KV cache.

Hamish Ivison, Yizhong Wang, Jiacheng Liu et al.

Learning from preference feedback has emerged as an essential step for improving the generation quality and performance of modern language models (LMs). Despite its widespread use, the way preference-based learning is applied varies wildly, with differing data, learning algorithms, and evaluations used, making disentangling the impact of each aspect difficult. In this work, we identify four core aspects of preference-based learning: preference data, learning algorithm, reward model, and policy training prompts, systematically investigate the impact of these components on downstream model performance, and suggest a recipe for strong learning for preference feedback. Our findings indicate that all aspects are important for performance, with better preference data leading to the largest improvements, followed by the choice of learning algorithm, the use of improved reward models, and finally the use of additional unlabeled prompts for policy training. Notably, PPO outperforms DPO by up to 2.5% in math and 1.2% in general domains. High-quality preference data leads to improvements of up to 8% in instruction following and truthfulness. Despite significant gains of up to 5% in mathematical evaluation when scaling up reward models, we surprisingly observe marginal improvements in other categories. We publicly release the code used for training (https://github.com/hamishivi/EasyLM) and evaluating (https://github.com/allenai/open-instruct) our models, along with the models and datasets themselves (https://huggingface.co/collections/allenai/tulu-v25-suite-66676520fd578080e126f618).

Seungone Kim, Juyoung Suk, Xiang Yue et al. · cmu

Given the increasing use of synthetic data in language model (LM) post-training, an LM's ability to generate high-quality data has become nearly as crucial as its ability to solve problems directly. While prior works have focused on developing effective data generation methods, they lack systematic comparison of different LMs as data generators in a unified setting. To address this gap, we propose AgoraBench, a benchmark that provides standardized settings and metrics to evaluate LMs' data generation abilities. Through synthesizing 1.26 million training instances using 6 LMs and training 99 student models, we uncover key insights about LMs' data generation capabilities. First, we observe that LMs exhibit distinct strengths. For instance, GPT-4o excels at generating new problems, while Claude-3.5-Sonnet performs better at enhancing existing ones. Furthermore, our analysis reveals that an LM's data generation ability doesn't necessarily correlate with its problem-solving ability. Instead, multiple intrinsic features of data quality-including response quality, perplexity, and instruction difficulty-collectively serve as better indicators. Finally, we demonstrate that strategic choices in output format and cost-conscious model selection significantly impact data generation effectiveness.

Wenhao Wu, Yizhong Wang, Yao Fu et al.

Effectively handling instructions with extremely long context remains a challenge for Large Language Models (LLMs), typically necessitating high-quality long data and substantial computational resources. This paper introduces Step-Skipping Alignment (SkipAlign), a new technique designed to enhance the long-context capabilities of LLMs in the phase of alignment without the need for additional efforts beyond training with original data length. SkipAlign is developed on the premise that long-range dependencies are fundamental to enhancing an LLM's capacity of long context. Departing from merely expanding the length of input samples, SkipAlign synthesizes long-range dependencies from the aspect of positions indices. This is achieved by the strategic insertion of skipped positions within instruction-following samples, which utilizes the semantic structure of the data to effectively expand the context. Through extensive experiments on base models with a variety of context window sizes, SkipAlign demonstrates its effectiveness across a spectrum of long-context tasks. Particularly noteworthy is that with a careful selection of the base model and alignment datasets, SkipAlign with only 6B parameters achieves it's best performance and comparable with strong baselines like GPT-3.5-Turbo-16K on LongBench.

Bowen Zhao, Zander Brumbaugh, Yizhong Wang et al. · uw

Language models (LMs) are trained on web text originating from many points in time and, in general, without any explicit temporal grounding. This work investigates the temporal chaos of pretrained LMs and explores various methods to align their internal knowledge to a target time, which we call "temporal alignment." To do this, we first automatically construct a dataset containing 20K time-sensitive questions and their answers for each year from 2000 to 2023. Based on this dataset, we empirically show that pretrained LMs (e.g., LLaMa2), despite having a recent pretraining cutoff (e.g., 2022), mostly answer questions using earlier knowledge (e.g., in 2019). We then develop several methods, from prompting to finetuning, to align LMs to use their most recent knowledge when answering questions, and investigate various factors in this alignment. Our experiments demonstrate that aligning LLaMa2 to the year 2022 can enhance its performance by up to 62% according to that year's answers. This improvement occurs even without explicitly mentioning time information, indicating the possibility of aligning models' internal sense of time after pretraining. Finally, we find that alignment to a historical time is also possible, with up to 2.8$\times$ the performance of the unaligned LM in 2010 if finetuning models to that year. These findings hint at the sophistication of LMs' internal knowledge organization and the necessity of tuning them properly.

Lester James V. Miranda, Yizhong Wang, Yanai Elazar et al. · allen-ai, cambridge

Learning from human feedback has enabled the alignment of language models (LMs) with human preferences. However, collecting human preferences is expensive and time-consuming, with highly variable annotation quality. An appealing alternative is to distill preferences from LMs as a source of synthetic annotations, offering a cost-effective and scalable alternative, albeit susceptible to other biases and errors. In this work, we introduce HyPER, a Hybrid Preference routER that defers an annotation to either humans or LMs, achieving better annotation quality while reducing the cost of human-only annotation. We formulate this as an optimization problem: given a preference dataset and an evaluation metric, we (1) train a performance prediction model (PPM) to predict a reward model's (RM) performance on an arbitrary combination of human and LM annotations and (2) employ a routing strategy that selects a combination that maximizes the predicted performance. We train the PPM on MultiPref, a new preference dataset with 10k instances paired with humans and LM labels. We show that the selected hybrid mixture of synthetic and direct human preferences using HyPER achieves better RM performance compared to using either one exclusively by 7-13% on RewardBench and generalizes across unseen preference datasets and other base models. We also observe the same trend in other benchmarks using Best-of-N reranking, where the hybrid mix has 2-3% better performance. Finally, we analyze features from HyPER and find that prompts with moderate safety concerns or complexity benefit the most from human feedback.

Qiyuan He, Yizhong Wang, Wenya Wang · uw

Large language models (LLMs) have demonstrated remarkable proficiency in understanding and generating responses to complex queries through large-scale pre-training. However, the efficacy of these models in memorizing and reasoning among large-scale structured knowledge, especially world knowledge that explicitly covers abundant factual information remains questionable. Addressing this gap, our research investigates whether LLMs can effectively store, recall, and reason with knowledge on a large scale comparable to latest knowledge bases (KBs) such as Wikidata. Specifically, we focus on three crucial aspects to study the viability: (1) the efficiency of LLMs with different sizes in memorizing the exact knowledge in the large-scale KB; (2) the flexibility of recalling the memorized knowledge in response to natural language queries; (3) the capability to infer new knowledge through reasoning. Our findings indicate that while LLMs hold promise as large-scale KBs capable of retrieving and responding with flexibility, enhancements in their reasoning capabilities are necessary to fully realize their potential.

Zhiyuan Zeng, Yizhong Wang, Hannaneh Hajishirzi et al. · uw

An ideal model evaluation should achieve two goals: identifying where the model fails and providing actionable improvement guidance. Toward these goals for language model (LM) evaluations, we formulate the problem of generating a weakness profile, a set of weaknesses expressed in natural language, given an LM's performance on every individual instance in a benchmark. We introduce a suite of quantitative assessments to compare different weakness profiling methods. We also introduce a weakness profiling method EvalTree. EvalTree constructs a capability tree where each node represents a capability described in natural language and is linked to a subset of benchmark instances that specifically evaluate this capability; it then extracts nodes where the LM performs poorly to generate a weakness profile. On the MATH and WildChat benchmarks, we show that EvalTree outperforms baseline weakness profiling methods by identifying weaknesses more precisely and comprehensively. Weakness profiling further enables weakness-guided data collection, and training data collection guided by EvalTree-identified weaknesses improves LM performance more than other data collection strategies. We also show how EvalTree exposes flaws in Chatbot Arena's human-voter-based evaluation practice. To facilitate future work, we provide an interface that allows practitioners to interactively explore the capability trees built by EvalTree.

Xinxi Lyu, Yizhong Wang, Hannaneh Hajishirzi et al.

Evaluating the capability of Large Language Models (LLMs) in following instructions has heavily relied on a powerful LLM as the judge, introducing unresolved biases that deviate the judgments from human judges. In this work, we reevaluate various choices for automatic evaluation on a wide range of instruction-following tasks. We experiment with methods that leverage human-written responses and observe that they enhance the reliability of automatic evaluations across a wide range of tasks, resulting in up to a 3.2% improvement in agreement with human judges. We also discovered that human-written responses offer an orthogonal perspective to model-generated responses in following instructions and should be used as an additional context when comparing model responses. Based on these observations, we develop a new evaluation benchmark, Human Response-Guided Evaluation of Instruction Following (HREF), comprising 4,258 samples across 11 task categories with a composite evaluation setup, employing a composite evaluation setup that selects the most reliable method for each category. In addition to providing reliable evaluation, HREF emphasizes individual task performance and is free from contamination. Finally, we study the impact of key design choices in HREF, including the size of the evaluation set, the judge model, the baseline model, and the prompt template. We host a live leaderboard that evaluates LLMs on the private evaluation set of HREF.

Rahul Nadkarni, Yizhong Wang, Noah A. Smith

Language model-based instruction-following systems have lately shown increasing performance on many benchmark tasks, demonstrating the capability of adapting to a broad variety of instructions. However, such systems are often not designed to be transparent about their limitations; a user may easily prompt a model with an instruction without any idea of whether the responses should be expected to be accurate, or if the system is even capable of performing the task. We propose a third party performance prediction framework, where a separate model is trained to predict the metric resulting from evaluating an instruction-following system on a task while assuming access only to its inputs and outputs at inference time. We perform this analysis with a variety of both open and closed instruction-following models as well as multiple performance predictors, and examine the effect of various factors such as model size, number of training tasks, and prompt format. Our findings indicate that third-party performance prediction is very challenging, and much work remains in developing predictors that can automatically reveal the limitations of modern instruction-following natural language processing systems.

Alisa Liu, Xiaochuang Han, Yizhong Wang et al.

Despite the general capabilities of large pretrained language models, they consistently benefit from further adaptation to better achieve desired behaviors. However, tuning these models has become increasingly resource-intensive, or impossible when model weights are private. We introduce proxy-tuning, a lightweight decoding-time algorithm that operates on top of black-box LMs to achieve the same end as direct tuning, but by accessing only its predictions over the output vocabulary, not its parameters. Our method tunes a smaller LM, then applies the difference between the predictions of the small tuned and untuned LMs to shift the original predictions of the larger untuned model in the direction of tuning, while retaining the benefits of larger-scale pretraining. In experiments, when we apply proxy-tuning to Llama2-70B using proxies of only 7B size, we can close 88% of the gap between Llama2-70B and its truly-tuned chat version, when evaluated across knowledge, reasoning, and safety benchmarks. We then demonstrate the generality of proxy-tuning by applying it to domain adaptation on code, and task-specific finetuning on question-answering and math problems. Finally, we show how to proxy-tune a truly black-box LM, GPT-3.5, for temporal adaptation, increasing its knowledge about recent events. Our work demonstrates the promise of using small tuned LMs to efficiently customize large, potentially proprietary LMs through decoding-time guidance.

Leo Z. Liu, Yizhong Wang, Jungo Kasai et al.

Models of language trained on very large corpora have been demonstrated useful for NLP. As fixed artifacts, they have become the object of intense study, with many researchers "probing" the extent to which linguistic abstractions, factual and commonsense knowledge, and reasoning abilities they acquire and readily demonstrate. Building on this line of work, we consider a new question: for types of knowledge a language model learns, when during (pre)training are they acquired? We plot probing performance across iterations, using RoBERTa as a case study. Among our findings: linguistic knowledge is acquired fast, stably, and robustly across domains. Facts and commonsense are slower and more domain-sensitive. Reasoning abilities are, in general, not stably acquired. As new datasets, pretraining protocols, and probes emerge, we believe that probing-across-time analyses can help researchers understand the complex, intermingled learning that these models undergo and guide us toward more efficient approaches that accomplish necessary learning faster.

Alon Talmor, Ori Yoran, Amnon Catav et al.

When answering complex questions, people can seamlessly combine information from visual, textual and tabular sources. While interest in models that reason over multiple pieces of evidence has surged in recent years, there has been relatively little work on question answering models that reason across multiple modalities. In this paper, we present MultiModalQA(MMQA): a challenging question answering dataset that requires joint reasoning over text, tables and images. We create MMQA using a new framework for generating complex multi-modal questions at scale, harvesting tables from Wikipedia, and attaching images and text paragraphs using entities that appear in each table. We then define a formal language that allows us to take questions that can be answered from a single modality, and combine them to generate cross-modal questions. Last, crowdsourcing workers take these automatically-generated questions and rephrase them into more fluent language. We create 29,918 questions through this procedure, and empirically demonstrate the necessity of a multi-modal multi-hop approach to solve our task: our multi-hop model, ImplicitDecomp, achieves an average F1of 51.7 over cross-modal questions, substantially outperforming a strong baseline that achieves 38.2 F1, but still lags significantly behind human performance, which is at 90.1 F1

Yue Guo, Wei Qiu, Yizhong Wang et al.

Health literacy has emerged as a crucial factor in making appropriate health decisions and ensuring treatment outcomes. However, medical jargon and the complex structure of professional language in this domain make health information especially hard to interpret. Thus, there is an urgent unmet need for automated methods to enhance the accessibility of the biomedical literature to the general population. This problem can be framed as a type of translation problem between the language of healthcare professionals, and that of the general public. In this paper, we introduce the novel task of automated generation of lay language summaries of biomedical scientific reviews, and construct a dataset to support the development and evaluation of automated methods through which to enhance the accessibility of the biomedical literature. We conduct analyses of the various challenges in solving this task, including not only summarization of the key points but also explanation of background knowledge and simplification of professional language. We experiment with state-of-the-art summarization models as well as several data augmentation techniques, and evaluate their performance using both automated metrics and human assessment. Results indicate that automatically generated summaries produced using contemporary neural architectures can achieve promising quality and readability as compared with reference summaries developed for the lay public by experts (best ROUGE-L of 50.24 and Flesch-Kincaid readability score of 13.30). We also discuss the limitations of the current attempt, providing insights and directions for future work.

Qianying Liu, Sicong Jiang, Yizhong Wang et al.

In this paper, we introduce LiveQA, a new question answering dataset constructed from play-by-play live broadcast. It contains 117k multiple-choice questions written by human commentators for over 1,670 NBA games, which are collected from the Chinese Hupu (https://nba.hupu.com/games) website. Derived from the characteristics of sports games, LiveQA can potentially test the reasoning ability across timeline-based live broadcasts, which is challenging compared to the existing datasets. In LiveQA, the questions require understanding the timeline, tracking events or doing mathematical computations. Our preliminary experiments show that the dataset introduces a challenging problem for question answering models, and a strong baseline model only achieves the accuracy of 53.1\% and cannot beat the dominant option rule. We release the code and data of this paper for future research.

Swabha Swayamdipta, Roy Schwartz, Nicholas Lourie et al.

Large datasets have become commonplace in NLP research. However, the increased emphasis on data quantity has made it challenging to assess the quality of data. We introduce Data Maps---a model-based tool to characterize and diagnose datasets. We leverage a largely ignored source of information: the behavior of the model on individual instances during training (training dynamics) for building data maps. This yields two intuitive measures for each example---the model's confidence in the true class, and the variability of this confidence across epochs---obtained in a single run of training. Experiments across four datasets show that these model-dependent measures reveal three distinct regions in the data map, each with pronounced characteristics. First, our data maps show the presence of "ambiguous" regions with respect to the model, which contribute the most towards out-of-distribution generalization. Second, the most populous regions in the data are "easy to learn" for the model, and play an important role in model optimization. Finally, data maps uncover a region with instances that the model finds "hard to learn"; these often correspond to labeling errors. Our results indicate that a shift in focus from quantity to quality of data could lead to robust models and improved out-of-distribution generalization.

Eric Wallace, Yizhong Wang, Sujian Li et al.

The ability to understand and work with numbers (numeracy) is critical for many complex reasoning tasks. Currently, most NLP models treat numbers in text in the same way as other tokens---they embed them as distributed vectors. Is this enough to capture numeracy? We begin by investigating the numerical reasoning capabilities of a state-of-the-art question answering model on the DROP dataset. We find this model excels on questions that require numerical reasoning, i.e., it already captures numeracy. To understand how this capability emerges, we probe token embedding methods (e.g., BERT, GloVe) on synthetic list maximum, number decoding, and addition tasks. A surprising degree of numeracy is naturally present in standard embeddings. For example, GloVe and word2vec accurately encode magnitude for numbers up to 1,000. Furthermore, character-level embeddings are even more precise---ELMo captures numeracy the best for all pre-trained methods---but BERT, which uses sub-word units, is less exact.

Xin Zhang, An Yang, Sujian Li et al.

Machine reading comprehension aims to teach machines to understand a text like a human and is a new challenging direction in Artificial Intelligence. This article summarizes recent advances in MRC, mainly focusing on two aspects (i.e., corpus and techniques). The specific characteristics of various MRC corpus are listed and compared. The main ideas of some typical MRC techniques are also described.

Dheeru Dua, Yizhong Wang, Pradeep Dasigi et al.

Reading comprehension has recently seen rapid progress, with systems matching humans on the most popular datasets for the task. However, a large body of work has highlighted the brittleness of these systems, showing that there is much work left to be done. We introduce a new English reading comprehension benchmark, DROP, which requires Discrete Reasoning Over the content of Paragraphs. In this crowdsourced, adversarially-created, 96k-question benchmark, a system must resolve references in a question, perhaps to multiple input positions, and perform discrete operations over them (such as addition, counting, or sorting). These operations require a much more comprehensive understanding of the content of paragraphs than what was necessary for prior datasets. We apply state-of-the-art methods from both the reading comprehension and semantic parsing literature on this dataset and show that the best systems only achieve 32.7% F1 on our generalized accuracy metric, while expert human performance is 96.0%. We additionally present a new model that combines reading comprehension methods with simple numerical reasoning to achieve 47.0% F1.

Yizhong Wang, Sujian Li, Jingfeng Yang

Discourse segmentation, which segments texts into Elementary Discourse Units, is a fundamental step in discourse analysis. Previous discourse segmenters rely on complicated hand-crafted features and are not practical in actual use. In this paper, we propose an end-to-end neural segmenter based on BiLSTM-CRF framework. To improve its accuracy, we address the problem of data insufficiency by transferring a word representation model that is trained on a large corpus. We also propose a restricted self-attention mechanism in order to capture useful information within a neighborhood. Experiments on the RST-DT corpus show that our model is significantly faster than previous methods, while achieving new state-of-the-art performance.

Shuming Ma, Xu Sun, Yizhong Wang et al.

A sentence can be translated into more than one correct sentences. However, most of the existing neural machine translation models only use one of the correct translations as the targets, and the other correct sentences are punished as the incorrect sentences in the training stage. Since most of the correct translations for one sentence share the similar bag-of-words, it is possible to distinguish the correct translations from the incorrect ones by the bag-of-words. In this paper, we propose an approach that uses both the sentences and the bag-of-words as targets in the training stage, in order to encourage the model to generate the potentially correct sentences that are not appeared in the training set. We evaluate our model on a Chinese-English translation dataset, and experiments show our model outperforms the strong baselines by the BLEU score of 4.55.

Yizhong Wang, Kai Liu, Jing Liu et al.

Machine reading comprehension (MRC) on real web data usually requires the machine to answer a question by analyzing multiple passages retrieved by search engine. Compared with MRC on a single passage, multi-passage MRC is more challenging, since we are likely to get multiple confusing answer candidates from different passages. To address this problem, we propose an end-to-end neural model that enables those answer candidates from different passages to verify each other based on their content representations. Specifically, we jointly train three modules that can predict the final answer based on three factors: the answer boundary, the answer content and the cross-passage answer verification. The experimental results show that our method outperforms the baseline by a large margin and achieves the state-of-the-art performance on the English MS-MARCO dataset and the Chinese DuReader dataset, both of which are designed for MRC in real-world settings.

Yizhong Wang, Sujian Li, Jingfeng Yang et al.

Identifying implicit discourse relations between text spans is a challenging task because it requires understanding the meaning of the text. To tackle this task, recent studies have tried several deep learning methods but few of them exploited the syntactic information. In this work, we explore the idea of incorporating syntactic parse tree into neural networks. Specifically, we employ the Tree-LSTM model and Tree-GRU model, which are based on the tree structure, to encode the arguments in a relation. Moreover, we further leverage the constituent tags to control the semantic composition process in these tree-structured neural networks. Experimental results show that our method achieves state-of-the-art performance on PDTB corpus.

Wei He, Kai Liu, Jing Liu et al.

This paper introduces DuReader, a new large-scale, open-domain Chinese ma- chine reading comprehension (MRC) dataset, designed to address real-world MRC. DuReader has three advantages over previous MRC datasets: (1) data sources: questions and documents are based on Baidu Search and Baidu Zhidao; answers are manually generated. (2) question types: it provides rich annotations for more question types, especially yes-no and opinion questions, that leaves more opportunity for the research community. (3) scale: it contains 200K questions, 420K answers and 1M documents; it is the largest Chinese MRC dataset so far. Experiments show that human performance is well above current state-of-the-art baseline systems, leaving plenty of room for the community to make improvements. To help the community make these improvements, both DuReader and baseline systems have been posted online. We also organize a shared competition to encourage the exploration of more models. Since the release of the task, there are significant improvements over the baselines.