Large Language Models (LLMs) for public use require continuous pre-training to remain up-to-date with the latest data. The models also need to be fine-tuned with specific instructions to maintain their ability to follow instructions accurately. Typically, LLMs are released in two versions: the Base LLM, pre-trained on diverse data, and the instruction-refined LLM, additionally trained with specific instructions for better instruction following. The question arises as to which model should undergo continuous pre-training to maintain its instruction-following abilities while also staying current with the latest data. In this study, we delve into the intricate relationship between continuous pre-training and instruction fine-tuning of the LLMs and investigate the impact of continuous pre-training on the instruction following abilities of both the base and its instruction finetuned model. Further, the instruction fine-tuning process is computationally intense and requires a substantial number of hand-annotated examples for the model to learn effectively. This study aims to find the most compute-efficient strategy to gain up-to-date knowledge and instruction-following capabilities without requiring any instruction data and fine-tuning. We empirically prove our findings on the LLaMa 3, 3.1 and Qwen 2, 2.5 family of base and instruction models, providing a comprehensive exploration of our hypotheses across varying sizes of pre-training data corpus and different LLMs settings.

Large Language Models (LLMs) have been adopted and deployed worldwide for a broad variety of applications. However, ensuring their safe use remains a significant challenge. Preference training and safety measures often overfit to harms prevalent in Western-centric datasets, and safety protocols frequently fail to extend to multilingual settings. In this work, we explore model merging in a diverse multi-task setting, combining safety and general-purpose tasks within a multilingual context. Each language introduces unique and varied learning challenges across tasks. We find that objective-based merging is more effective than mixing data, with improvements of up to 8% and 10% in general performance and safety respectively. We also find that language-based merging is highly effective -- by merging monolingually fine-tuned models, we achieve a 4% increase in general performance and 7% reduction in harm across all languages on top of the data mixtures method using the same available data. Overall, our comprehensive study of merging approaches provides a useful framework for building strong and safe multilingual models.

In this paper, we investigate the safety mechanisms of instruction fine-tuned large language models (LLMs). We discover that re-weighting MLP neurons can significantly compromise a model's safety, especially for MLPs in end-of-sentence inferences. We hypothesize that LLMs evaluate the harmfulness of prompts during end-of-sentence inferences, and MLP layers plays a critical role in this process. Based on this hypothesis, we develop 2 novel white-box jailbreak methods: a prompt-specific method and a prompt-general method. The prompt-specific method targets individual prompts and optimizes the attack on the fly, while the prompt-general method is pre-trained offline and can generalize to unseen harmful prompts. Our methods demonstrate robust performance across 7 popular opensource LLMs, size ranging from 2B to 72B. Furthermore, our study provides insights into vulnerabilities of instruction-tuned LLM's safety and deepens the understanding of the internal mechanisms of LLMs. The capabilities of large language models (LLMs) have improved rapidly in recent years (Achiam et al., 2023; Anthropic, 2023; Touvron et al., 2023). One of the primary ways of deploying LLMs in practice is through chatbots. Instruction fine-tuning is the most common approach for transforming a pre-trained LLM into an effective chatbot (Wei et al., 2021; Ouyang et al., 2022; Chung et al., 2022). This process involves training the model on a variety of prompt-response pairs, marked with special tokens, to guide the model in following instructions and generating helpful, relevant responses.

Recent large language model (LLM)-driven chat assistant systems have integrated memory components to track user-assistant chat histories, enabling more accurate and personalized responses. However, their long-term memory capabilities in sustained interactions remain underexplored. This paper introduces LongMemEval, a comprehensive benchmark designed to evaluate five core long-term memory abilities of chat assistants: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention. With 500 meticulously curated questions embedded within freely scalable user-assistant chat histories, LongMemEval presents a significant challenge to existing long-term memory systems, with commercial chat assistants and long-context LLMs showing 30% accuracy drop on memorizing information across sustained interactions. We then present a unified framework that breaks down the long-term memory design into four design choices across the indexing, retrieval, and reading stages. Built upon key experimental insights, we propose several memory designs including session decomposition for optimizing value granularity, fact-augmented key expansion for enhancing the index structure, and time-aware query expansion for refining the search scope. Experiment results show that these optimizations greatly improve both memory recall and downstream question answering on LongMemEval. Overall, our study provides valuable resources and guidance for advancing the long-term memory capabilities of LLM-based chat assistants, paving the way toward more personalized and reliable conversational AI.

Multilingual Large Language Models (LLMs) have demonstrated exceptional performance in Machine Translation (MT) tasks. However, their MT abilities in the context of code-switching (the practice of mixing two or more languages in an utterance) remain under-explored. In this paper, we introduce Rule-Based Prompting, a novel prompting technique to generate code-mixed sentences. We measure and compare the code-mixed MT abilities of 3 popular multilingual LLMs: GPT-3.5-turbo, GPT-4, and Gemini Pro across five language pairs: English-{Hindi, Bengali, Gujarati, French, Spanish} using $k$-shot prompting ($k\in\{0, 1, 10, 20\}$) and Rule-Based Prompting. Our findings suggest that though $k$-shot prompting often leads to the best results, Rule-Based prompting shows promise in generating unique code-mixed sentences that vary in their style of code-mixing. We also use $k$-shot prompting to gauge the code-mixed to English translation abilities of multilingual LLMs. For this purpose, we create a gold-standard code-mixed dataset spanning five language pairs: English-{Hindi, Bengali, Gujarati, French, Spanish}. As a real-world application of our work, we create a code-mixed chatbot.

Recent studies have shown that LLMs are vulnerable to denial-of-service (DoS) attacks, where adversarial inputs like spelling errors or non-semantic prompts trigger endless outputs without generating an [EOS] token. These attacks can potentially cause high latency and make LLM services inaccessible to other users or tasks. However, when there are speech-to-text interfaces (e.g., voice commands to a robot), executing such DoS attacks becomes challenging, as it is difficult to introduce spelling errors or non-semantic prompts through speech. A simple DoS attack in these scenarios would be to instruct the model to "Keep repeating Hello", but we observe that relying solely on natural instructions limits output length, which is bounded by the maximum length of the LLM's supervised finetuning (SFT) data. To overcome this limitation, we propose poisoning-based DoS (P-DoS) attacks for LLMs, demonstrating that injecting a single poisoned sample designed for DoS purposes can break the output length limit. For example, a poisoned sample can successfully attack GPT-4o and GPT-4o mini (via OpenAI's finetuning API) using less than $1, causing repeated outputs up to the maximum inference length (16K tokens, compared to 0.5K before poisoning). Additionally, we perform comprehensive ablation studies on open-source LLMs and extend our method to LLM agents, where attackers can control both the finetuning dataset and algorithm. Our findings underscore the urgent need for defenses against P-DoS attacks to secure LLMs. Our code is available at https://github.com/sail-sg/P-DoS.

Open-generation bias benchmarks evaluate social biases in Large Language Models (LLMs) by analyzing their outputs. However, the classifiers used in analysis often have inherent biases, leading to unfair conclusions. This study examines such biases in open-generation benchmarks like BOLD and SAGED. Using the MGSD dataset, we conduct two experiments. The first uses counterfactuals to measure prediction variations across demographic groups by altering stereotype-related prefixes. The second applies explainability tools (SHAP) to validate that the observed biases stem from these counterfactuals. Results reveal unequal treatment of demographic descriptors, calling for more robust bias metric models.

The generation and editing of floor plans are critical in architectural planning, requiring a high degree of flexibility and efficiency. Existing methods demand extensive input information and lack the capability for interactive adaptation to user modifications. This paper introduces ChatHouseDiffusion, which leverages large language models (LLMs) to interpret natural language input, employs graphormer to encode topological relationships, and uses diffusion models to flexibly generate and edit floor plans. This approach allows iterative design adjustments based on user ideas, significantly enhancing design efficiency. Compared to existing models, ChatHouseDiffusion achieves higher Intersection over Union (IoU) scores, permitting precise, localized adjustments without the need for complete redesigns, thus offering greater practicality. Experiments demonstrate that our model not only strictly adheres to user specifications but also facilitates a more intuitive design process through its interactive capabilities.

Session-based recommendation is the task of predicting the next item a user will interact with, often without access to historical user data. In this work, we introduce Sequential Masked Modeling, a novel approach for encoder-only transformer architectures to tackle the challenges of single-session recommendation. Our method combines data augmentation through window sliding with a unique penultimate token masking strategy to capture sequential dependencies more effectively. By enhancing how transformers handle session data, Sequential Masked Modeling significantly improves next-item prediction performance. We evaluate our approach on three widely-used datasets, Yoochoose 1/64, Diginetica, and Tmall, comparing it to state-of-the-art single-session, cross-session, and multi-relation approaches. The results demonstrate that our Transformer-SMM models consistently outperform all models that rely on the same amount of information, while even rivaling methods that have access to more extensive user history. This study highlights the potential of encoder-only transformers in session-based recommendation and opens the door for further improvements.

Code-mixing, the practice of alternating between two or more languages in an utterance, is a common phenomenon in multilingual communities. Due to the colloquial nature of code-mixing, there is no singular correct way to translate an English sentence into a code-mixed sentence. For this reason, standard n-gram-based MT evaluation metrics such as the BLEU score are not appropriate for code-mixed evaluation. To demonstrate this, we propose a novel method for code-mixed text generation: Controlled Generation, which parameterizes the code-mixing degree (CMD) and enables the generation of multiple semantically equivalent code-mixed sentences from a given English sentence. We introduce a robust new evaluation metric: GAME: A Gold-Standard Agnostic Measure for Evaluation of Code-Mixed Sentences. GAME is both language-agnostic and gold-standard-agnostic, i.e. unlike other metrics, GAME does not require gold-standard code-mixed sentences for evaluation, thus eliminating the need for human annotators in the code-mixed evaluation process. When used to evaluate semantically equivalent code-mixed sentences, we find that GAME scores have a lower standard deviation than BLEU scores. Further, we create and release a dataset containing gold-standard code-mixed sentences across 4 language pairs: English-{Hindi, Bengali, French, Spanish} to encourage more computational research on code-mixing.