We introduce the LiLiuM series of large language models (LLMs): 1B, 7B, and 13B parameter models developed 100% in-house to fit eBay's specific needs in the e-commerce domain. This gives eBay full control over all aspects of the models including license, data, vocabulary, and architecture. We expect these models to be used as a foundation for fine-tuning and instruction-tuning, eliminating dependencies to external models. The LiLiuM LLMs have been trained on 3 trillion tokens of multilingual text from general and e-commerce domain. They perform similar to the popular LLaMA-2 models on English natural language understanding (NLU) benchmarks. At the same time, we outperform LLaMA-2 on non-English NLU tasks, machine translation and on e-commerce specific downstream tasks. As part of our data mixture, we utilize the newly released RedPajama-V2 dataset for training and share our insights regarding data filtering and deduplication. We also discuss in detail how to serialize structured data for use in autoregressive language modeling. We provide insights on the effects of including code and parallel machine translation data in pre-training. Furthermore, we develop our own tokenizer and model vocabulary, customized towards e-commerce. This way, we can achieve up to 34% speed-up in text generation on eBay-specific downstream tasks compared to LLaMA-2. Finally, in relation to LLM pretraining, we show that checkpoint averaging can further improve over the best individual model checkpoint.
In the context of real-world applications, leveraging large language models (LLMs) for domain-specific tasks often faces two major challenges: domain-specific knowledge privacy and constrained resources. To address these issues, we propose PDSS, a privacy-preserving framework for step-by-step distillation of LLMs. PDSS works on a server-client architecture, wherein client transmits perturbed prompts to the server's LLM for rationale generation. The generated rationales are then decoded by the client and used to enrich the training of task-specific small language model(SLM) within a multi-task learning paradigm. PDSS introduces two privacy protection strategies: the Exponential Mechanism Strategy and the Encoder-Decoder Strategy, balancing prompt privacy and rationale usability. Experiments demonstrate the effectiveness of PDSS in various text generation tasks, enabling the training of task-specific SLM with enhanced performance while prioritizing data privacy protection.
Low-rank adaptation (LoRA) has become the default approach to fine-tune large language models (LLMs) due to its significant reduction in trainable parameters. However, trainable parameter demand for LoRA increases with increasing model embedding dimensions, leading to high compute costs. Additionally, its backward updates require storing high-dimensional intermediate activations and optimizer states, demanding high peak GPU memory. In this paper, we introduce large model fine-tuning via spectrally decomposed low-dimensional adaptation (LaMDA), a novel approach to fine-tuning large language models, which leverages low-dimensional adaptation to achieve significant reductions in trainable parameters and peak GPU memory footprint. LaMDA freezes a first projection matrix (PMA) in the adaptation path while introducing a low-dimensional trainable square matrix, resulting in substantial reductions in trainable parameters and peak GPU memory usage. LaMDA gradually freezes a second projection matrix (PMB) during the early fine-tuning stages, reducing the compute cost associated with weight updates to enhance parameter efficiency further. We also present an enhancement, LaMDA++, incorporating a ``lite-weight" adaptive rank allocation for the LoRA path via normalized spectrum analysis of pre-trained model weights. We evaluate LaMDA/LaMDA++ across various tasks, including natural language understanding with the GLUE benchmark, text summarization, natural language generation, and complex reasoning on different LLMs. Results show that LaMDA matches or surpasses the performance of existing alternatives while requiring up to 17.7x fewer parameter updates and up to 1.32x lower peak GPU memory usage during fine-tuning. Code will be publicly available.
Dynamic text-attributed graphs (DyTAGs) are prevalent in various real-world scenarios, where each node and edge are associated with text descriptions, and both the graph structure and text descriptions evolve over time. Despite their broad applicability, there is a notable scarcity of benchmark datasets tailored to DyTAGs, which hinders the potential advancement in many research fields. To address this gap, we introduce Dynamic Text-attributed Graph Benchmark (DTGB), a collection of large-scale, time-evolving graphs from diverse domains, with nodes and edges enriched by dynamically changing text attributes and categories. To facilitate the use of DTGB, we design standardized evaluation procedures based on four real-world use cases: future link prediction, destination node retrieval, edge classification, and textual relation generation. These tasks require models to understand both dynamic graph structures and natural language, highlighting the unique challenges posed by DyTAGs. Moreover, we conduct extensive benchmark experiments on DTGB, evaluating 7 popular dynamic graph learning algorithms and their variants of adapting to text attributes with LLM embeddings, along with 6 powerful large language models (LLMs). Our results show the limitations of existing models in handling DyTAGs. Our analysis also demonstrates the utility of DTGB in investigating the incorporation of structural and textual dynamics. The proposed DTGB fosters research on DyTAGs and their broad applications. It offers a comprehensive benchmark for evaluating and advancing models to handle the interplay between dynamic graph structures and natural language. The dataset and source code are available at https://github.com/zjs123/DTGB.
In scenarios where language models must incorporate new information efficiently without extensive retraining, traditional fine-tuning methods are prone to overfitting, degraded generalization, and unnatural language generation. To address these limitations, we introduce Consistent In-Context Editing (ICE), a novel approach leveraging the model's in-context learning capability to optimize toward a contextual distribution rather than a one-hot target. ICE introduces a simple yet effective optimization framework for the model to internalize new knowledge by aligning its output distributions with and without additional context. This method enhances the robustness and effectiveness of gradient-based tuning methods, preventing overfitting and preserving the model's integrity. We analyze ICE across four critical aspects of knowledge editing: accuracy, locality, generalization, and linguistic quality, demonstrating its advantages. Experimental results confirm the effectiveness of ICE and demonstrate its potential for continual editing, ensuring that the integrity of the model is preserved while updating information.
Large Language Models (LLMs) possess vast amounts of knowledge within their parameters, prompting research into methods for locating and editing this knowledge. Previous work has largely focused on locating entity-related (often single-token) facts in smaller models. However, several key questions remain unanswered: (1) How can we effectively locate query-relevant neurons in decoder-only LLMs, such as Llama and Mistral? (2) How can we address the challenge of long-form (or free-form) text generation? (3) Are there localized knowledge regions in LLMs? In this study, we introduce Query-Relevant Neuron Cluster Attribution (QRNCA), a novel architecture-agnostic framework capable of identifying query-relevant neurons in LLMs. QRNCA allows for the examination of long-form answers beyond triplet facts by employing the proxy task of multi-choice question answering. To evaluate the effectiveness of our detected neurons, we build two multi-choice QA datasets spanning diverse domains and languages. Empirical evaluations demonstrate that our method outperforms baseline methods significantly. Further, analysis of neuron distributions reveals the presence of visible localized regions, particularly within different domains. Finally, we show potential applications of our detected neurons in knowledge editing and neuron-based prediction.
Table-to-text generation, a long-standing challenge in natural language generation, has remained unexplored through the lens of subjectivity. Subjectivity here encompasses the comprehension of information derived from the table that cannot be described solely by objective data. Given the absence of pre-existing datasets, we introduce the Ta2TS dataset with 3849 data instances. We perform the task of fine-tuning sequence-to-sequence models on the linearized tables and prompting on popular large language models. We analyze the results from a quantitative and qualitative perspective to ensure the capture of subjectivity and factual consistency. The analysis shows the fine-tuned LMs can perform close to the prompted LLMs. Both the models can capture the tabular data, generating texts with 85.15% BERTScore and 26.28% Meteor score. To the best of our knowledge, we provide the first-of-its-kind dataset on tables with multiple genres and subjectivity included and present the first comprehensive analysis and comparison of different LLM performances on this task.
Productive interactions between diverse users and language technologies require outputs from the latter to be culturally relevant and sensitive. Prior works have evaluated models' knowledge of cultural norms, values, and artifacts, without considering how this knowledge manifests in downstream applications. In this work, we focus on extrinsic evaluation of cultural competence in two text generation tasks, open-ended question answering and story generation. We quantitatively and qualitatively evaluate model outputs when an explicit cue of culture, specifically nationality, is perturbed in the prompts. Although we find that model outputs do vary when varying nationalities and feature culturally relevant words, we also find weak correlations between text similarity of outputs for different countries and the cultural values of these countries. Finally, we discuss important considerations in designing comprehensive evaluation of cultural competence in user-facing tasks.
We present a method for rewriting an input sentence to match specific values of nontrivial linguistic features, such as dependency depth. In contrast to earlier work, our method uses in-context learning rather than finetuning, making it applicable in use cases where data is sparse. We show that our model performs accurate rewrites and matches the state of the art on rewriting sentences to a specified school grade level.
Large Language Models (LLMs) like OpenAI's GPT series, Anthropic's Claude, and Meta's LLaMa have shown remarkable capabilities in text generation. However, their susceptibility to toxic prompts presents significant security challenges. This paper investigates alignment techniques, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), to mitigate these risks. We conduct an empirical study on refusal patterns across nine LLMs, revealing that models with uniform refusal patterns, such as Claude3, exhibit higher security. Based on these findings, we propose self-distilling and cross-model distilling methods to enhance LLM security. Our results show that these methods significantly improve refusal rates and reduce unsafe content, with cross-model distilling achieving refusal rates close to Claude3's 94.51%. These findings underscore the potential of distillation-based alignment in securing LLMs against toxic prompts.
Interleaved image-text generation has emerged as a crucial multimodal task, aiming at creating sequences of interleaved visual and textual content given a query. Despite notable advancements in recent multimodal large language models (MLLMs), generating integrated image-text sequences that exhibit narrative coherence and entity and style consistency remains challenging due to poor training data quality. To address this gap, we introduce CoMM, a high-quality Coherent interleaved image-text MultiModal dataset designed to enhance the coherence, consistency, and alignment of generated multimodal content. Initially, CoMM harnesses raw data from diverse sources, focusing on instructional content and visual storytelling, establishing a foundation for coherent and consistent content. To further refine the data quality, we devise a multi-perspective filter strategy that leverages advanced pre-trained models to ensure the development of sentences, consistency of inserted images, and semantic alignment between them. Various quality evaluation metrics are designed to prove the high quality of the filtered dataset. Meanwhile, extensive few-shot experiments on various downstream tasks demonstrate CoMM's effectiveness in significantly enhancing the in-context learning capabilities of MLLMs. Moreover, we propose four new tasks to evaluate MLLMs' interleaved generation abilities, supported by a comprehensive evaluation framework. We believe CoMM opens a new avenue for advanced MLLMs with superior multimodal in-context learning and understanding ability.
Despite the striking advances in recent language generation performance, model-generated responses have suffered from the chronic problem of hallucinations that are either untrue or unfaithful to a given source. Especially in the task of knowledge grounded conversation, the models are required to generate informative responses, but hallucinated utterances lead to miscommunication. In particular, entity-level hallucination that causes critical misinformation and undesirable conversation is one of the major concerns. To address this issue, we propose a post-hoc refinement method called REM. It aims to enhance the quality and faithfulness of hallucinated utterances by refining them based on the source knowledge. If the generated utterance has a low source-faithfulness score with the given knowledge, REM mines the key entities in the knowledge and implicitly uses them for refining the utterances. We verify that our method reduces entity hallucination in the utterance. Also, we show the adaptability and efficacy of REM with extensive experiments and generative results. Our code is available at https://github.com/YOONNAJANG/REM.
This paper introduces CodeGemma, a collection of specialized open code models built on top of Gemma, capable of a variety of code and natural language generation tasks. We release three model variants. CodeGemma 7B pretrained (PT) and instruction-tuned (IT) variants have remarkably resilient natural language understanding, excel in mathematical reasoning, and match code capabilities of other open models. CodeGemma 2B is a state-of-the-art code completion model designed for fast code infilling and open-ended generation in latency-sensitive settings.
Large language models (LLMs) have shown promising abilities as cost-effective and reference-free evaluators for assessing language generation quality. In particular, pairwise LLM evaluators, which compare two generated texts and determine the preferred one, have been employed in a wide range of applications. However, LLMs exhibit preference biases and worrying sensitivity to prompt designs. In this work, we first reveal that the predictive preference of LLMs can be highly brittle and skewed, even with semantically equivalent instructions. We find that fairer predictive preferences from LLMs consistently lead to judgments that are better aligned with humans. Motivated by this phenomenon, we propose an automatic Zero-shot Evaluation-oriented Prompt Optimization framework, ZEPO, which aims to produce fairer preference decisions and improve the alignment of LLM evaluators with human judgments. To this end, we propose a zero-shot learning objective based on the preference decision fairness. ZEPO demonstrates substantial performance improvements over state-of-the-art LLM evaluators, without requiring labeled data, on representative meta-evaluation benchmarks. Our findings underscore the critical correlation between preference fairness and human alignment, positioning ZEPO as an efficient prompt optimizer for bridging the gap between LLM evaluators and human judgments.
While the success of large language models (LLMs) increases demand for machine-generated text, current pay-per-token pricing schemes create a misalignment of incentives known in economics as moral hazard: Text-generating agents have strong incentive to cut costs by preferring a cheaper model over the cutting-edge one, and this can be done "behind the scenes" since the agent performs inference internally. In this work, we approach this issue from an economic perspective, by proposing a pay-for-performance, contract-based framework for incentivizing quality. We study a principal-agent game where the agent generates text using costly inference, and the contract determines the principal's payment for the text according to an automated quality evaluation. Since standard contract theory is inapplicable when internal inference costs are unknown, we introduce cost-robust contracts. As our main theoretical contribution, we characterize optimal cost-robust contracts through a direct correspondence to optimal composite hypothesis tests from statistics, generalizing a result of Saig et al. (NeurIPS'23). We evaluate our framework empirically by deriving contracts for a range of objectives and LLM evaluation benchmarks, and find that cost-robust contracts sacrifice only a marginal increase in objective value compared to their cost-aware counterparts.
The advent of black-box deep neural network classification models has sparked the need to explain their decisions. However, in the case of generative AI, such as large language models (LLMs), there is no class prediction to explain. Rather, one can ask why an LLM output a particular response to a given prompt. In this paper, we answer this question by proposing, to the best of our knowledge, the first contrastive explanation methods requiring simply black-box/query access. Our explanations suggest that an LLM outputs a reply to a given prompt because if the prompt was slightly modified, the LLM would have given a different response that is either less preferable or contradicts the original response. The key insight is that contrastive explanations simply require a scoring function that has meaning to the user and not necessarily a specific real valued quantity (viz. class label). We offer two algorithms for finding contrastive explanations: i) A myopic algorithm, which although effective in creating contrasts, requires many model calls and ii) A budgeted algorithm, our main algorithmic contribution, which intelligently creates contrasts adhering to a query budget, necessary for longer contexts. We show the efficacy of these methods on diverse natural language tasks such as open-text generation, automated red teaming, and explaining conversational degradation.
Elicitation interviews are the most common requirements elicitation technique, and proficiency in conducting these interviews is crucial for requirements elicitation. Traditional training methods, typically limited to textbook learning, may not sufficiently address the practical complexities of interviewing techniques. Practical training with various interview scenarios is important for understanding how to apply theoretical knowledge in real-world contexts. However, there is a shortage of educational interview material, as creating interview scripts requires both technical expertise and creativity. To address this issue, we develop a specialized GPT agent for auto-generating interview scripts. The GPT agent is equipped with a dedicated knowledge base tailored to the guidelines and best practices of requirements elicitation interview procedures. We employ a prompt chaining approach to mitigate the output length constraint of GPT to be able to generate thorough and detailed interview scripts. This involves dividing the interview into sections and crafting distinct prompts for each, allowing for the generation of complete content for each section. The generated scripts are assessed through standard natural language generation evaluation metrics and an expert judgment study, confirming their applicability in requirements engineering training.
The relationship between the quality of a string, as judged by a human reader, and its probability, $p(\boldsymbol{y})$ under a language model undergirds the development of better language models. For example, many popular algorithms for sampling from a language model have been conceived with the goal of manipulating $p(\boldsymbol{y})$ to place higher probability on strings that humans deem of high quality. In this article, we examine the probability--quality relationship in language models explicitly aligned to human preferences, e.g., through reinforcement learning through human feedback. We show that, when sampling corpora from an aligned language model, there exists a trade-off between the strings' average reward and average log-likelihood under the prior language model, i.e., the same model before alignment with human preferences. We provide a formal treatment of this phenomenon and demonstrate how a choice of sampling adaptor allows for a selection of how much likelihood we exchange for the reward.
This research investigates prompt designs of evaluating generated texts using large language models (LLMs). While LLMs are increasingly used for scoring various inputs, creating effective prompts for open-ended text evaluation remains challenging due to model sensitivity and subjectivity in evaluation of text generation. Our study experimented with different prompt structures, altering the sequence of output instructions and including explanatory reasons. We found that the order of presenting reasons and scores significantly influences LLMs' scoring, with a different level of rule understanding in the prompt. An additional optimization may enhance scoring alignment if sufficient data is available. This insight is crucial for improving the accuracy and consistency of LLM-based evaluations.
Controllable text generation (CTG) seeks to craft texts adhering to specific attributes, traditionally employing learning-based techniques such as training, fine-tuning, or prefix-tuning with attribute-specific datasets. These approaches, while effective, demand extensive computational and data resources. In contrast, some proposed learning-free alternatives circumvent learning but often yield inferior results, exemplifying the fundamental machine learning trade-off between computational expense and model efficacy. To overcome these limitations, we propose FreeCtrl, a learning-free approach that dynamically adjusts the weights of selected feedforward neural network (FFN) vectors to steer the outputs of large language models (LLMs). FreeCtrl hinges on the principle that the weights of different FFN vectors influence the likelihood of different tokens appearing in the output. By identifying and adaptively adjusting the weights of attribute-related FFN vectors, FreeCtrl can control the output likelihood of attribute keywords in the generated content. Extensive experiments on single- and multi-attribute control reveal that the learning-free FreeCtrl outperforms other learning-free and learning-based methods, successfully resolving the dilemma between learning costs and model performance.