Generations from large language models (LLMs) can be improved by sampling and scoring multiple solutions to select a final answer. Current "sample and select" methods such as self-consistency (SC) rely on majority voting to score answers. However, when tasks have many distinct and valid answers, selection by voting requires a large number of samples. This makes SC prohibitively expensive for interactive tasks that involve generating multiple actions (answers) sequentially. After establishing that majority voting fails to provide consistent gains on such tasks, we demonstrate how to increase success rates by softening the scoring criterion. We introduce Soft Self-Consistency (Soft-SC), which replaces SC's discontinuous scoring with a continuous score computed from model likelihoods, allowing for selection even when actions are sparsely distributed. Soft-SC improves both performance and efficiency on long-horizon interactive tasks, requiring half as many samples as SC for comparable or better performance. For a fixed number of samples, Soft-SC leads to a 1.3% increase over SC in absolute success rate on writing bash programs, a 6.6% increase on online shopping (WebShop), and a 4.7% increase for an interactive household game (ALFWorld). Finally, we show that Soft-SC can be applied to both open-source and black-box models.
While large language models (LLMs) are increasingly being used for program synthesis, they lack the global view needed to develop useful abstractions; they generally predict programs one at a time, often repeating the same functionality. Generating redundant code from scratch is both inefficient and error-prone. To address this, we propose Refactoring for Generalizable Abstraction Learning (ReGAL), a gradient-free method for learning a library of reusable functions via code refactorization, i.e. restructuring code without changing its execution output. ReGAL learns from a small set of existing programs, iteratively verifying and refining its abstractions via execution. We find that the shared function libraries discovered by ReGAL make programs easier to predict across diverse domains. On three datasets (LOGO graphics generation, Date reasoning, and TextCraft, a Minecraft-based text game), both open-source and proprietary LLMs improve in accuracy when predicting programs with ReGAL functions. For CodeLlama-13B, ReGAL results in absolute accuracy increases of 11.5% on graphics, 26.1% on date understanding, and 8.1% on TextCraft, outperforming GPT-3.5 in two of three domains. Our analysis reveals ReGAL's abstractions encapsulate frequently-used subroutines as well as environment dynamics.
Large Language Models (LLMs) are increasingly being used for interactive decision-making tasks requiring planning and adapting to the environment. Recent works employ LLMs-as-agents in broadly two ways: iteratively determining the next action (iterative executors) or generating plans and executing sub-tasks using LLMs (plan-and-execute). However, these methods struggle with task complexity, as the inability to execute any sub-task may lead to task failure. To address these shortcomings, we introduce As-Needed Decomposition and Planning for complex Tasks (ADaPT), an approach that explicitly plans and decomposes complex sub-tasks as-needed, i.e., when the LLM is unable to execute them. ADaPT recursively decomposes sub-tasks to adapt to both task complexity and LLM capability. Our results demonstrate that ADaPT substantially outperforms established strong baselines, achieving success rates up to 28.3% higher in ALFWorld, 27% in WebShop, and 33% in TextCraft -- a novel compositional dataset that we introduce. Through extensive analysis, we illustrate the importance of multilevel decomposition and establish that ADaPT dynamically adjusts to the capabilities of the executor LLM as well as to task complexity.
An increasing number of vision-language tasks can be handled with little to no training, i.e., in a zero and few-shot manner, by marrying large language models (LLMs) to vision encoders, resulting in large vision-language models (LVLMs). While this has huge upsides, such as not requiring training data or custom architectures, how an input is presented to a LVLM can have a major impact on zero-shot model performance. In particular, inputs phrased in an underspecified way can result in incorrect answers due to factors like missing visual information, complex implicit reasoning, or linguistic ambiguity. Therefore, adding visually grounded information to the input as a preemptive clarification should improve model performance by reducing underspecification, e.g., by localizing objects and disambiguating references. Similarly, in the VQA setting, changing the way questions are framed can make them easier for models to answer. To this end, we present Rephrase, Augment and Reason (RepARe), a gradient-free framework that extracts salient details about the image using the underlying LVLM as a captioner and reasoner, in order to propose modifications to the original question. We then use the LVLM's confidence over a generated answer as an unsupervised scoring function to select the rephrased question most likely to improve zero-shot performance. Focusing on two visual question answering tasks, we show that RepARe can result in a 3.85% (absolute) increase in zero-shot performance on VQAv2 and a 6.41% point increase on A-OKVQA. Additionally, we find that using gold answers for oracle question candidate selection achieves a substantial gain in VQA accuracy by up to 14.41%. Through extensive analysis, we demonstrate that outputs from RepARe increase syntactic complexity, and effectively utilize vision-language interaction and the frozen language model in LVLMs.
Multi-step reasoning ability is fundamental to many natural language tasks, yet it is unclear what constitutes a good reasoning chain and how to evaluate them. Most existing methods focus solely on whether the reasoning chain leads to the correct conclusion, but this answer-oriented view may confound the quality of reasoning with other spurious shortcuts to predict the answer. To bridge this gap, we evaluate reasoning chains by viewing them as informal proofs that derive the final answer. Specifically, we propose ReCEval (Reasoning Chain Evaluation), a framework that evaluates reasoning chains through two key properties: (1) correctness, i.e., each step makes a valid inference based on the information contained within the step, preceding steps, and input context, and (2) informativeness, i.e., each step provides new information that is helpful towards deriving the generated answer. We implement ReCEval using natural language inference models and information-theoretic measures. On multiple datasets, ReCEval is highly effective in identifying different types of errors, resulting in notable improvements compared to prior methods. We demonstrate that our informativeness metric captures the expected flow of information in high-quality reasoning chains and we also analyze the impact of previous steps on evaluating correctness and informativeness. Finally, we show that scoring reasoning chains based on ReCEval can improve downstream performance of reasoning tasks. Our code is publicly available at: https://github.com/archiki/ReCEval
Providing natural language instructions in prompts is a useful new paradigm for improving task performance of large language models in a zero-shot setting. Recent work has aimed to improve such prompts via manual rewriting or gradient-based tuning. However, manual rewriting is time-consuming and requires subjective interpretation, while gradient-based tuning can be extremely computationally demanding for large models and requires full access to model weights, which may not be available for API-based models. In this work, we introduce Gradient-free Instructional Prompt Search (GrIPS), a gradient-free, edit-based search approach for improving task instructions for large language models. GrIPS takes in instructions designed for humans and automatically returns an improved, edited prompt, while allowing for API-based tuning. The instructions in our search are iteratively edited using four operations (delete, add, swap, paraphrase) on text at the phrase-level. With InstructGPT models, GrIPS improves the average task performance by up to 4.30 percentage points on eight classification tasks from the Natural-Instructions dataset. We see improvements for both instruction-only prompts and for k-shot example+instruction prompts. Notably, GrIPS outperforms manual rewriting following the guidelines in Mishra et al. (2022) and also outperforms purely example-based prompts while controlling for the available compute and data budget. Lastly, we provide qualitative analysis of the edited instructions across several scales of GPT models. Our code is available at: https://github.com/archiki/GrIPS
While recent benchmarks have spurred a lot of new work on improving the generalization of pretrained multilingual language models on multilingual tasks, techniques to improve code-switched natural language understanding tasks have been far less explored. In this work, we propose the use of bilingual intermediate pretraining as a reliable technique to derive large and consistent performance gains on three different NLP tasks using code-switched text. We achieve substantial absolute improvements of 7.87%, 20.15%, and 10.99%, on the mean accuracies and F1 scores over previous state-of-the-art systems for Hindi-English Natural Language Inference (NLI), Question Answering (QA) tasks, and Spanish-English Sentiment Analysis (SA) respectively. We show consistent performance gains on four different code-switched language-pairs (Hindi-English, Spanish-English, Tamil-English and Malayalam-English) for SA. We also present a code-switched masked language modelling (MLM) pretraining technique that consistently benefits SA compared to standard MLM pretraining using real code-switched text.
End-to-end models for robust automatic speech recognition (ASR) have not been sufficiently well-explored in prior work. With end-to-end models, one could choose to preprocess the input speech using speech enhancement techniques and train the model using enhanced speech. Another alternative is to pass the noisy speech as input and modify the model architecture to adapt to noisy speech. A systematic comparison of these two approaches for end-to-end robust ASR has not been attempted before. We address this gap and present a detailed comparison of speech enhancement-based techniques and three different model-based adaptation techniques covering data augmentation, multi-task learning, and adversarial learning for robust ASR. While adversarial learning is the best-performing technique on certain noise types, it comes at the cost of degrading clean speech WER. On other relatively stationary noise types, a new speech enhancement technique outperformed all the model-based adaptation techniques. This suggests that knowledge of the underlying noise type can meaningfully inform the choice of adaptation technique.