Samson Tan

BigScience Workshop, Teven Le Scao, Angela Fan et al. · allen-ai, berkeley

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License.

Shiqi Wang, Zheng Li, Haifeng Qian et al. · amazon-science, ibm-research

Code generation models have achieved impressive performance. However, they tend to be brittle as slight edits to a prompt could lead to very different generations; these robustness properties, critical for user experience when deployed in real-life applications, are not well understood. Most existing works on robustness in text or code tasks have focused on classification, while robustness in generation tasks is an uncharted area and to date there is no comprehensive benchmark for robustness in code generation. In this paper, we propose ReCode, a comprehensive robustness evaluation benchmark for code generation models. We customize over 30 transformations specifically for code on docstrings, function and variable names, code syntax, and code format. They are carefully designed to be natural in real-life coding practice, preserve the original semantic meaning, and thus provide multifaceted assessments of a model's robustness performance. With human annotators, we verified that over 90% of the perturbed prompts do not alter the semantic meaning of the original prompt. In addition, we define robustness metrics for code generation models considering the worst-case behavior under each type of perturbation, taking advantage of the fact that executing the generated code can serve as objective evaluation. We demonstrate ReCode on SOTA models using HumanEval, MBPP, as well as function completion tasks derived from them. Interesting observations include: better robustness for CodeGen over InCoder and GPT-J; models are most sensitive to syntax perturbations; more challenging robustness evaluation on MBPP over HumanEval.

Yacine Jernite, Huu Nguyen, Stella Biderman et al. · allen-ai, cmu

The recent emergence and adoption of Machine Learning technology, and specifically of Large Language Models, has drawn attention to the need for systematic and transparent management of language data. This work proposes an approach to global language data governance that attempts to organize data management amongst stakeholders, values, and rights. Our proposal is informed by prior work on distributed governance that accounts for human values and grounded by an international research collaboration that brings together researchers and practitioners from 60 countries. The framework we present is a multi-party international governance structure focused on language data, and incorporating technical and organizational tools needed to support its work.

Tuan Dinh, Jinman Zhao, Samson Tan et al. · amazon-science

Large language models of code (Code-LLMs) have recently brought tremendous advances to code completion, a fundamental feature of programming assistance and code intelligence. However, most existing works ignore the possible presence of bugs in the code context for generation, which are inevitable in software development. Therefore, we introduce and study the buggy-code completion problem, inspired by the realistic scenario of real-time code suggestion where the code context contains potential bugs -- anti-patterns that can become bugs in the completed program. To systematically study the task, we introduce two datasets: one with synthetic bugs derived from semantics-altering operator changes (buggy-HumanEval) and one with realistic bugs derived from user submissions to coding problems (buggy-FixEval). We find that the presence of potential bugs significantly degrades the generation performance of the high-performing Code-LLMs. For instance, the passing rates of CODEGEN-2B-MONO on test cases of buggy-HumanEval drop more than 50% given a single potential bug in the context. Finally, we investigate several post-hoc methods for mitigating the adverse effect of potential bugs and find that there remains a significant gap in post-mitigation performance.

Amazon AGI, Aaron Langford, Aayush Shah et al. · amazon-science

We present Amazon Nova, a new generation of state-of-the-art foundation models that deliver frontier intelligence and industry-leading price performance. Amazon Nova Pro is a highly-capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks. Amazon Nova Lite is a low-cost multimodal model that is lightning fast for processing images, video, documents and text. Amazon Nova Micro is a text-only model that delivers our lowest-latency responses at very low cost. Amazon Nova Canvas is an image generation model that creates professional grade images with rich customization controls. Amazon Nova Reel is a video generation model offering high-quality outputs, customization, and motion control. Our models were built responsibly and with a commitment to customer trust, security, and reliability. We report benchmarking results for core capabilities, agentic performance, long context, functional adaptation, runtime performance, and human evaluation.

Guangsen Wang, Samson Tan, Shafiq Joty et al.

We present BotSIM, a data-efficient end-to-end Bot SIMulation toolkit for commercial text-based task-oriented dialog (TOD) systems. BotSIM consists of three major components: 1) a Generator that can infer semantic-level dialog acts and entities from bot definitions and generate user queries via model-based paraphrasing; 2) an agenda-based dialog user Simulator (ABUS) to simulate conversations with the dialog agents; 3) a Remediator to analyze the simulated conversations, visualize the bot health reports and provide actionable remediation suggestions for bot troubleshooting and improvement. We demonstrate BotSIM's effectiveness in end-to-end evaluation, remediation and multi-intent dialog generation via case studies on two commercial bot platforms. BotSIM's "generation-simulation-remediation" paradigm accelerates the end-to-end bot evaluation and iteration process by: 1) reducing manual test cases creation efforts; 2) enabling a holistic gauge of the bot in terms of NLU and end-to-end performance via extensive dialog simulation; 3) improving the bot troubleshooting process with actionable suggestions. A demo of our system can be found at https://tinyurl.com/mryu74cd and a demo video at https://youtu.be/qLi5iSoly30. We have open-sourced the toolkit at https://github.com/salesforce/botsim

Hengchang Hu, Yiming Cao, Zhankui He et al.

Fairness is a widely discussed topic in recommender systems, but its practical implementation faces challenges in defining sensitive features while maintaining recommendation accuracy. We propose feature fairness as the foundation to achieve equitable treatment across diverse groups defined by various feature combinations. This improves overall accuracy through balanced feature generalizability. We introduce unbiased feature learning through adversarial training, using adversarial perturbation to enhance feature representation. The adversaries improve model generalization for under-represented features. We adapt adversaries automatically based on two forms of feature biases: frequency and combination variety of feature values. This allows us to dynamically adjust perturbation strengths and adversarial training weights. Stronger perturbations are applied to feature values with fewer combination varieties to improve generalization, while higher weights for low-frequency features address training imbalances. We leverage the Adaptive Adversarial perturbation based on the widely-applied Factorization Machine (AAFM) as our backbone model. In experiments, AAFM surpasses strong baselines in both fairness and accuracy measures. AAFM excels in providing item- and user-fairness for single- and multi-feature tasks, showcasing their versatility and scalability. To maintain good accuracy, we find that adversarial perturbation must be well-managed: during training, perturbations should not overly persist and their strengths should decay.

Samson Tan, Araz Taeihagh, Kathy Baxter

The speed and scale at which machine learning (ML) systems are deployed are accelerating even as an increasing number of studies highlight their potential for negative impact. There is a clear need for companies and regulators to manage the risk from proposed ML systems before they harm people. To achieve this, private and public sector actors first need to identify the risks posed by a proposed ML system. A system's overall risk is influenced by its direct and indirect effects. However, existing frameworks for ML risk/impact assessment often address an abstract notion of risk or do not concretize this dependence. We propose to address this gap with a context-sensitive framework for identifying ML system risks comprising two components: a taxonomy of the first- and second-order risks posed by ML systems, and their contributing factors. First-order risks stem from aspects of the ML system, while second-order risks stem from the consequences of first-order risks. These consequences are system failures that result from design and development choices. We explore how different risks may manifest in various types of ML systems, the factors that affect each risk, and how first-order risks may lead to second-order effects when the system interacts with the real world. Throughout the paper, we show how real events and prior research fit into our Machine Learning System Risk framework (MLSR). MLSR operates on ML systems rather than technologies or domains, recognizing that a system's design, implementation, and use case all contribute to its risk. In doing so, it unifies the risks that are commonly discussed in the ethical AI community (e.g., ethical/human rights risks) with system-level risks (e.g., application, design, control risks), paving the way for holistic risk assessments of ML systems.

Bo Ai, Yuchen Wang, Yugin Tan et al.

Authorship attribution is the task of identifying the author of a given text. The key is finding representations that can differentiate between authors. Existing approaches typically use manually designed features that capture a dataset's content and style, but these approaches are dataset-dependent and yield inconsistent performance across corpora. In this work, we propose \textit{learning} author-specific representations by fine-tuning pre-trained generic language representations with a contrastive objective (Contra-X). We show that Contra-X learns representations that form highly separable clusters for different authors. It advances the state-of-the-art on multiple human and machine authorship attribution benchmarks, enabling improvements of up to 6.8% over cross-entropy fine-tuning. However, we find that Contra-X improves overall accuracy at the cost of sacrificing performance for some authors. Resolving this tension will be an important direction for future work. To the best of our knowledge, we are the first to integrate contrastive learning with pre-trained language model fine-tuning for authorship attribution.

Stella Biderman, Hailey Schoelkopf, Lintang Sutawika et al. · cmu

Effective evaluation of language models remains an open challenge in NLP. Researchers and engineers face methodological issues such as the sensitivity of models to evaluation setup, difficulty of proper comparisons across methods, and the lack of reproducibility and transparency. In this paper we draw on three years of experience in evaluating large language models to provide guidance and lessons for researchers. First, we provide an overview of common challenges faced in language model evaluation. Second, we delineate best practices for addressing or lessening the impact of these challenges on research. Third, we present the Language Model Evaluation Harness (lm-eval): an open source library for independent, reproducible, and extensible evaluation of language models that seeks to address these issues. We describe the features of the library as well as case studies in which the library has been used to alleviate these methodological concerns.

Kaustubh D. Dhole, Varun Gangal, Sebastian Gehrmann et al.

Data augmentation is an important component in the robustness evaluation of models in natural language processing (NLP) and in enhancing the diversity of the data they are trained on. In this paper, we present NL-Augmenter, a new participatory Python-based natural language augmentation framework which supports the creation of both transformations (modifications to the data) and filters (data splits according to specific features). We describe the framework and an initial set of 117 transformations and 23 filters for a variety of natural language tasks. We demonstrate the efficacy of NL-Augmenter by using several of its transformations to analyze the robustness of popular natural language models. The infrastructure, datacards and robustness analysis results are available publicly on the NL-Augmenter repository (https://github.com/GEM-benchmark/NL-Augmenter).

Vyas Raina, Samson Tan, Volkan Cevher et al.

Deep learning-based Natural Language Processing (NLP) models are vulnerable to adversarial attacks, where small perturbations can cause a model to misclassify. Adversarial Training (AT) is often used to increase model robustness. However, we have discovered an intriguing phenomenon: deliberately or accidentally miscalibrating models masks gradients in a way that interferes with adversarial attack search methods, giving rise to an apparent increase in robustness. We show that this observed gain in robustness is an illusion of robustness (IOR), and demonstrate how an adversary can perform various forms of test-time temperature calibration to nullify the aforementioned interference and allow the adversarial attack to find adversarial examples. Hence, we urge the NLP community to incorporate test-time temperature scaling into their robustness evaluations to ensure that any observed gains are genuine. Finally, we show how the temperature can be scaled during \textit{training} to improve genuine robustness.

Tobias Falke, Nicolas Anastassacos, Samson Tan et al.

Sparse Mixture-of-Experts (MoE) architectures are increasingly popular for frontier large language models (LLM) but they introduce training challenges due to routing complexity. Fully leveraging parameters of an MoE model requires all experts to be well-trained and to specialize in non-redundant ways. Assessing this, however, is complicated due to lack of established metrics and, importantly, many routing techniques exhibit similar performance at smaller sizes, which is often not reflective of their behavior at large scale. To address this challenge, we propose the MoE Routing Testbed, a setup that gives clearer visibility into routing dynamics at small scale while using realistic data. The testbed pairs a data mix with clearly distinguishable domains with a reference router that prescribes ideal routing based on these domains, providing a well-defined upper bound for comparison. This enables quantifiable measurement of expert specialization. To demonstrate the value of the testbed, we compare various MoE routing approaches and show that balancing scope is the crucial factor that allows specialization while maintaining high expert utilization. We confirm that this observation generalizes to models 35x larger.

Rami Aly, Zhiqiang Tang, Samson Tan et al.

Large Language Models (LLMs) frequently hallucinate, impeding their reliability in mission-critical situations. One approach to address this issue is to provide citations to relevant sources alongside generated content, enhancing the verifiability of generations. However, citing passages accurately in answers remains a substantial challenge. This paper proposes a weakly-supervised fine-tuning method leveraging factual consistency models (FCMs). Our approach alternates between generating texts with citations and supervised fine-tuning with FCM-filtered citation data. Focused learning is integrated into the objective, directing the fine-tuning process to emphasise the factual unit tokens, as measured by an FCM. Results on the ALCE few-shot citation benchmark with various instruction-tuned LLMs demonstrate superior performance compared to in-context learning, vanilla supervised fine-tuning, and state-of-the-art methods, with an average improvement of $34.1$, $15.5$, and $10.5$ citation F$_1$ points, respectively. Moreover, in a domain transfer setting we show that the obtained citation generation ability robustly transfers to unseen datasets. Notably, our citation improvements contribute to the lowest factual error rate across baselines.

Sabrina J. Mielke, Zaid Alyafeai, Elizabeth Salesky et al.

What are the units of text that we want to model? From bytes to multi-word expressions, text can be analyzed and generated at many granularities. Until recently, most natural language processing (NLP) models operated over words, treating those as discrete and atomic tokens, but starting with byte-pair encoding (BPE), subword-based approaches have become dominant in many areas, enabling small vocabularies while still allowing for fast inference. Is the end of the road character-level model or byte-level processing? In this survey, we connect several lines of work from the pre-neural and neural era, by showing how hybrid approaches of words and characters as well as subword-based approaches based on learned segmentation have been proposed and evaluated. We conclude that there is and likely will never be a silver bullet singular solution for all applications and that thinking seriously about tokenization remains important for many applications.

Keng Ji Chow, Samson Tan, Min-Yen Kan

Numerous visio-linguistic (V+L) representation learning methods have been developed, yet existing datasets do not adequately evaluate the extent to which they represent visual and linguistic concepts in a unified space. We propose several novel evaluation settings for V+L models, including cross-modal transfer. Furthermore, existing V+L benchmarks often report global accuracy scores on the entire dataset, making it difficult to pinpoint the specific reasoning tasks that models fail and succeed at. We present TraVLR, a synthetic dataset comprising four V+L reasoning tasks. TraVLR's synthetic nature allows us to constrain its training and testing distributions along task-relevant dimensions, enabling the evaluation of out-of-distribution generalisation. Each example in TraVLR redundantly encodes the scene in two modalities, allowing either to be dropped or added during training or testing without losing relevant information. We compare the performance of four state-of-the-art V+L models, finding that while they perform well on test examples from the same modality, they all fail at cross-modal transfer and have limited success accommodating the addition or deletion of one modality. We release TraVLR as an open challenge for the research community.

Yunxiang Zhang, Liangming Pan, Samson Tan et al.

Modern Natural Language Processing (NLP) models are known to be sensitive to input perturbations and their performance can decrease when applied to real-world, noisy data. However, it is still unclear why models are less robust to some perturbations than others. In this work, we test the hypothesis that the extent to which a model is affected by an unseen textual perturbation (robustness) can be explained by the learnability of the perturbation (defined as how well the model learns to identify the perturbation with a small amount of evidence). We further give a causal justification for the learnability metric. We conduct extensive experiments with four prominent NLP models -- TextRNN, BERT, RoBERTa and XLNet -- over eight types of textual perturbations on three datasets. We show that a model which is better at identifying a perturbation (higher learnability) becomes worse at ignoring such a perturbation at test time (lower robustness), providing empirical support for our hypothesis.

Samson Tan, Shafiq Joty, Kathy Baxter et al.

Questions of fairness, robustness, and transparency are paramount to address before deploying NLP systems. Central to these concerns is the question of reliability: Can NLP systems reliably treat different demographics fairly and function correctly in diverse and noisy environments? To address this, we argue for the need for reliability testing and contextualize it among existing work on improving accountability. We show how adversarial attacks can be reframed for this goal, via a framework for developing reliability tests. We argue that reliability testing -- with an emphasis on interdisciplinary collaboration -- will enable rigorous and targeted testing, and aid in the enactment and enforcement of industry standards.

Samson Tan, Shafiq Joty

Multilingual models have demonstrated impressive cross-lingual transfer performance. However, test sets like XNLI are monolingual at the example level. In multilingual communities, it is common for polyglots to code-mix when conversing with each other. Inspired by this phenomenon, we present two strong black-box adversarial attacks (one word-level, one phrase-level) for multilingual models that push their ability to handle code-mixed sentences to the limit. The former uses bilingual dictionaries to propose perturbations and translations of the clean example for sense disambiguation. The latter directly aligns the clean example with its translations before extracting phrases as perturbations. Our phrase-level attack has a success rate of 89.75% against XLM-R-large, bringing its average accuracy of 79.85 down to 8.18 on XNLI. Finally, we propose an efficient adversarial training scheme that trains in the same number of steps as the original model and show that it improves model accuracy.

Karan Goel, Nazneen Rajani, Jesse Vig et al.

Despite impressive performance on standard benchmarks, deep neural networks are often brittle when deployed in real-world systems. Consequently, recent research has focused on testing the robustness of such models, resulting in a diverse set of evaluation methodologies ranging from adversarial attacks to rule-based data transformations. In this work, we identify challenges with evaluating NLP systems and propose a solution in the form of Robustness Gym (RG), a simple and extensible evaluation toolkit that unifies 4 standard evaluation paradigms: subpopulations, transformations, evaluation sets, and adversarial attacks. By providing a common platform for evaluation, Robustness Gym enables practitioners to compare results from all 4 evaluation paradigms with just a few clicks, and to easily develop and share novel evaluation methods using a built-in set of abstractions. To validate Robustness Gym's utility to practitioners, we conducted a real-world case study with a sentiment-modeling team, revealing performance degradations of 18%+. To verify that Robustness Gym can aid novel research analyses, we perform the first study of state-of-the-art commercial and academic named entity linking (NEL) systems, as well as a fine-grained analysis of state-of-the-art summarization models. For NEL, commercial systems struggle to link rare entities and lag their academic counterparts by 10%+, while state-of-the-art summarization models struggle on examples that require abstraction and distillation, degrading by 9%+. Robustness Gym can be found at https://robustnessgym.com/

Samson Tan, Shafiq Joty, Min-Yen Kan et al.

Training on only perfect Standard English corpora predisposes pre-trained neural networks to discriminate against minorities from non-standard linguistic backgrounds (e.g., African American Vernacular English, Colloquial Singapore English, etc.). We perturb the inflectional morphology of words to craft plausible and semantically similar adversarial examples that expose these biases in popular NLP models, e.g., BERT and Transformer, and show that adversarially fine-tuning them for a single epoch significantly improves robustness without sacrificing performance on clean data.

Samson Tan, Shafiq Joty, Lav R. Varshney et al.

Inflectional variation is a common feature of World Englishes such as Colloquial Singapore English and African American Vernacular English. Although comprehension by human readers is usually unimpaired by non-standard inflections, current NLP systems are not yet robust. We propose Base-Inflection Encoding (BITE), a method to tokenize English text by reducing inflected words to their base forms before reinjecting the grammatical information as special symbols. Fine-tuning pretrained NLP models for downstream tasks using our encoding defends against inflectional adversaries while maintaining performance on clean data. Models using BITE generalize better to dialects with non-standard inflections without explicit training and translation models converge faster when trained with BITE. Finally, we show that our encoding improves the vocabulary efficiency of popular data-driven subword tokenizers. Since there has been no prior work on quantitatively evaluating vocabulary efficiency, we propose metrics to do so.