Learning to Generate Instruction Tuning Datasets for Zero-Shot Task Adaptation

We introduce Bonito, an open-source model for conditional task generation: the task of converting unannotated text into task-specific training datasets for instruction tuning. Our goal is to enable zero-shot task adaptation of large language models on users' specialized, private data. We train Bonito on a new large-scale dataset with 1.65M examples created by remixing existing instruction tuning datasets into meta-templates. The meta-templates for a dataset produce training examples where the input is the unannotated text and the task attribute and the output consists of the instruction and the response. We use Bonito to generate synthetic tasks for seven datasets from specialized domains across three task types -- yes-no question answering, extractive question answering, and natural language inference -- and adapt language models. We show that Bonito significantly improves the average performance of pretrained and instruction tuned models over the de facto self supervised baseline. For example, adapting Mistral-Instruct-v2 and instruction tuned variants of Mistral and Llama2 with Bonito improves the strong zero-shot performance by 22.1 F1 points whereas the next word prediction objective undoes some of the benefits of instruction tuning and reduces the average performance by 0.8 F1 points. We conduct additional experiments with Bonito to understand the effects of the domain, the size of the training set, and the choice of alternative synthetic task generators. Overall, we show that learning with synthetic instruction tuning datasets is an effective way to adapt language models to new domains. The model, dataset, and code are available at https://github.com/BatsResearch/bonito.

CEREAL: Few-Sample Clustering Evaluation

Evaluating clustering quality with reliable evaluation metrics like normalized mutual information (NMI) requires labeled data that can be expensive to annotate. We focus on the underexplored problem of estimating clustering quality with limited labels. We adapt existing approaches from the few-sample model evaluation literature to actively sub-sample, with a learned surrogate model, the most informative data points for annotation to estimate the evaluation metric. However, we find that their estimation can be biased and only relies on the labeled data. To that end, we introduce CEREAL, a comprehensive framework for few-sample clustering evaluation that extends active sampling approaches in three key ways. First, we propose novel NMI-based acquisition functions that account for the distinctive properties of clustering and uncertainties from a learned surrogate model. Next, we use ideas from semi-supervised learning and train the surrogate model with both the labeled and unlabeled data. Finally, we pseudo-label the unlabeled data with the surrogate model. We run experiments to estimate NMI in an active sampling pipeline on three datasets across vision and language. Our results show that CEREAL reduces the area under the absolute error curve by up to 57% compared to the best sampling baseline. We perform an extensive ablation study to show that our framework is agnostic to the choice of clustering algorithm and evaluation metric. We also extend CEREAL from clusterwise annotations to pairwise annotations. Overall, CEREAL can efficiently evaluate clustering with limited human annotations.

Learning to Compose Soft Prompts for Compositional Zero-Shot Learning

We introduce compositional soft prompting (CSP), a parameter-efficient learning technique to improve the zero-shot compositionality of large-scale pretrained vision-language models (VLMs) without the overhead of fine-tuning the entire model. VLMs can represent arbitrary classes as natural language prompts in their flexible text encoders but they underperform state-of-the-art methods on compositional zero-shot benchmark tasks. To improve VLMs, we propose a novel form of soft prompting. We treat the attributes and objects that are composed to define classes as learnable tokens of vocabulary and tune them on multiple prompt compositions. During inference, we recompose the learned attribute-object vocabulary in new combinations and show that CSP outperforms the original VLM on benchmark datasets by an average of 14.7 percentage points of accuracy. CSP also achieves new state-of-the-art accuracies on two out of three benchmark datasets, while only fine-tuning a small number of parameters. Further, we show that CSP improves generalization to higher-order attribute-attribute-object compositions and combinations of pretrained attributes and fine-tuned objects.

PromptSource: An Integrated Development Environment and Repository for Natural Language Prompts

PromptSource is a system for creating, sharing, and using natural language prompts. Prompts are functions that map an example from a dataset to a natural language input and target output. Using prompts to train and query language models is an emerging area in NLP that requires new tools that let users develop and refine these prompts collaboratively. PromptSource addresses the emergent challenges in this new setting with (1) a templating language for defining data-linked prompts, (2) an interface that lets users quickly iterate on prompt development by observing outputs of their prompts on many examples, and (3) a community-driven set of guidelines for contributing new prompts to a common pool. Over 2,000 prompts for roughly 170 datasets are already available in PromptSource. PromptSource is available at https://github.com/bigscience-workshop/promptsource.

TAGLETS: A System for Automatic Semi-Supervised Learning with Auxiliary Data

Machine learning practitioners often have access to a spectrum of data: labeled data for the target task (which is often limited), unlabeled data, and auxiliary data, the many available labeled datasets for other tasks. We describe TAGLETS, a system built to study techniques for automatically exploiting all three types of data and creating high-quality, servable classifiers. The key components of TAGLETS are: (1) auxiliary data organized according to a knowledge graph, (2) modules encapsulating different methods for exploiting auxiliary and unlabeled data, and (3) a distillation stage in which the ensembled modules are combined into a servable model. We compare TAGLETS with state-of-the-art transfer learning and semi-supervised learning methods on four image classification tasks. Our study covers a range of settings, varying the amount of labeled data and the semantic relatedness of the auxiliary data to the target task. We find that the intelligent incorporation of auxiliary and unlabeled data into multiple learning techniques enables TAGLETS to match-and most often significantly surpass-these alternatives. TAGLETS is available as an open-source system at github.com/BatsResearch/taglets.

Zero-Shot Learning with Common Sense Knowledge Graphs

Zero-shot learning relies on semantic class representations such as attributes or pretrained embeddings to predict classes without any labeled examples. We propose to learn class representations from common sense knowledge graphs. Common sense knowledge graphs are an untapped source of explicit high-level knowledge that requires little human effort to apply to a range of tasks. To capture the knowledge in the graph, we introduce ZSL-KG, a framework based on graph neural networks with non-linear aggregators to generate class representations. Whereas most prior work on graph neural networks uses linear functions to aggregate information from neighboring nodes, we find that non-linear aggregators such as LSTMs or transformers lead to significant improvements on zero-shot tasks. On two natural language tasks across three datasets, ZSL-KG shows an average improvement of 9.2 points of accuracy versus state-of-the-art methods. In addition, on an object classification task, ZSL-KG shows a 2.2 accuracy point improvement versus the best methods that do not require hand-engineered class representations. Finally, we find that ZSL-KG outperforms the best performing graph neural networks with linear aggregators by an average of 3.8 points of accuracy across these four datasets.