Large Language Models (LLMs) have transformed natural language processing, but face significant challenges in widespread deployment due to their high runtime cost. In this paper, we introduce SeedLM, a novel post-training compression method that uses seeds of pseudo-random generators to encode and compress model weights. Specifically, for each block of weights, we find a seed that is fed into a Linear Feedback Shift Register (LFSR) during inference to efficiently generate a random matrix. This matrix is then linearly combined with compressed coefficients to reconstruct the weight block. SeedLM reduces memory access and leverages idle compute cycles during inference, effectively speeding up memory-bound tasks by trading compute for fewer memory accesses. Unlike state-of-the-art compression methods that rely on calibration data, our approach is data-free and generalizes well across diverse tasks. Our experiments with Llama 3 70B, which is particularly challenging to compress, show that SeedLM achieves significantly better zero-shot accuracy retention at 4- and 3-bit than state-of-the-art techniques, while maintaining performance comparable to FP16 baselines. Additionally, FPGA-based tests demonstrate that 4-bit SeedLM, as model size increases to 70B, approaches a 4x speed-up over an FP16 Llama 2/3 baseline.

The development of Large Language Models (LLMs) has significantly advanced various AI applications in commercial and scientific research fields, such as scientific literature summarization, writing assistance, and knowledge graph construction. However, a significant challenge is the high risk of hallucination during LLM inference, which can lead to security concerns like factual inaccuracies, inconsistent information, and fabricated content. To tackle this issue, it is essential to develop effective methods for reducing hallucination while maintaining the original capabilities of the LLM. This paper introduces a novel approach called Iterative Model-level Contrastive Learning (Iter-AHMCL) to address hallucination. This method modifies the representation layers of pre-trained LLMs by using contrastive `positive' and `negative' models, trained on data with and without hallucinations. By leveraging the differences between these two models, we create a more straightforward pathway to eliminate hallucinations, and the iterative nature of contrastive learning further enhances performance. Experimental validation on four pre-trained foundation LLMs (LLaMA2, Alpaca, LLaMA3, and Qwen) finetuning with a specially designed dataset shows that our approach achieves an average improvement of 10.1 points on the TruthfulQA benchmark. Comprehensive experiments demonstrate the effectiveness of Iter-AHMCL in reducing hallucination while maintaining the general capabilities of LLMs.

While Large language model (LLM)-based programming assistants such as CoPilot and ChatGPT can help improve the productivity of professional software developers, they can also facilitate cheating in introductory computer programming courses. Assuming instructors have limited control over the industrial-strength models, this paper investigates the baseline performance of 5 widely used LLMs on a collection of introductory programming problems, examines adversarial perturbations to degrade their performance, and describes the results of a user study aimed at understanding the efficacy of such perturbations in hindering actual code generation for introductory programming assignments. The user study suggests that i) perturbations combinedly reduced the average correctness score by 77%, ii) the drop in correctness caused by these perturbations was affected based on their detectability.

Large language models (LLMs) have become integral tool for users from various backgrounds. LLMs, trained on vast corpora, reflect the linguistic and cultural nuances embedded in their pre-training data. However, the values and perspectives inherent in this data can influence the behavior of LLMs, leading to potential biases. As a result, the use of LLMs in contexts involving spiritual or moral values necessitates careful consideration of these underlying biases. Our work starts with verification of our hypothesis by testing the spiritual values of popular LLMs. Experimental results show that LLMs' spiritual values are quite diverse, as opposed to the stereotype of atheists or secularists. We then investigate how different spiritual values affect LLMs in social-fairness scenarios e.g., hate speech identification). Our findings reveal that different spiritual values indeed lead to different sensitivity to different hate target groups. Furthermore, we propose to continue pre-training LLMs on spiritual texts, and empirical results demonstrate the effectiveness of this approach in mitigating spiritual bias.

In-context learning has been recognized as a key factor in the success of Large Language Models (LLMs). It refers to the model's ability to learn patterns on the fly from provided in-context examples in the prompt during inference. Previous studies have demonstrated that the Transformer architecture used in LLMs can implement a single-step gradient descent update by processing in-context examples in a single forward pass. Recent work has further shown that, during in-context learning, a looped Transformer can implement multi-step gradient descent updates in forward passes. However, their theoretical results require an exponential number of in-context examples, $n = \exp(\Omega(T))$, where $T$ is the number of loops or passes, to achieve a reasonably low error. In this paper, we study linear looped Transformers in-context learning on linear vector generation tasks. We show that linear looped Transformers can implement multi-step gradient descent efficiently for in-context learning. Our results demonstrate that as long as the input data has a constant condition number, e.g., $n = O(d)$, the linear looped Transformers can achieve a small error by multi-step gradient descent during in-context learning. Furthermore, our preliminary experiments validate our theoretical analysis. Our findings reveal that the Transformer architecture possesses a stronger in-context learning capability than previously understood, offering new insights into the mechanisms behind LLMs and potentially guiding the better design of efficient inference algorithms for LLMs.

Large language models (LLMs) typically improve performance by either retrieving semantically similar information, or enhancing reasoning abilities through structured prompts like chain-of-thought. While both strategies are considered crucial, it remains unclear which has a greater impact on model performance or whether a combination of both is necessary. This paper answers this question by proposing a knowledge graph (KG)-based random-walk reasoning approach that leverages causal relationships. We conduct experiments on the commonsense question answering task that is based on a KG. The KG inherently provides both relevant information, such as related entity keywords, and a reasoning structure through the connections between nodes. Experimental results show that the proposed KG-based random-walk reasoning method improves the reasoning ability and performance of LLMs. Interestingly, incorporating three seemingly irrelevant sentences into the query using KG-based random-walk reasoning enhances LLM performance, contrary to conventional wisdom. These findings suggest that integrating causal structures into prompts can significantly improve reasoning capabilities, providing new insights into the role of causality in optimizing LLM performance.

Quantization has been substantially adopted to accelerate inference and reduce memory consumption of large language models (LLMs). While activation-weight joint quantization speeds up the inference process through low-precision kernels, we demonstrate that it suffers severe performance degradation on multi-step reasoning tasks, rendering it ineffective. We propose a novel quantization paradigm called QSPEC, which seamlessly integrates two complementary quantization schemes for speculative decoding. Leveraging nearly cost-free execution switching, QSPEC drafts tokens with low-precision, fast activation-weight quantization, and verifies them with high-precision weight-only quantization, effectively combining the strengths of both quantization schemes. Compared to high-precision quantization methods, QSPEC empirically boosts token generation throughput by up to 1.80x without any quality compromise, distinguishing it from other low-precision quantization approaches. This enhancement is also consistent across various serving tasks, model sizes, quantization methods, and batch sizes. Unlike existing speculative decoding techniques, our approach reuses weights and the KV cache, avoiding additional memory overhead. Furthermore, QSPEC offers a plug-and-play advantage without requiring any training. We believe that QSPEC demonstrates unique strengths for future deployment of high-fidelity quantization schemes, particularly in memory-constrained scenarios (e.g., edge devices).

Large language models (LLMs) have exhibited outstanding performance in engaging with humans and addressing complex questions by leveraging their vast implicit knowledge and robust reasoning capabilities. However, such models are vulnerable to jailbreak attacks, leading to the generation of harmful responses. Despite recent research on single-turn jailbreak strategies to facilitate the development of defence mechanisms, the challenge of revealing vulnerabilities under multi-turn setting remains relatively under-explored. In this work, we propose Jigsaw Puzzles (JSP), a straightforward yet effective multi-turn jailbreak strategy against the advanced LLMs. JSP splits questions into harmless fractions as the input of each turn, and requests LLMs to reconstruct and respond to questions under multi-turn interaction. Our experimental results demonstrate that the proposed JSP jailbreak bypasses original safeguards against explicitly harmful content, achieving an average attack success rate of 93.76% on 189 harmful queries across 5 advanced LLMs (Gemini-1.5-Pro, Llama-3.1-70B, GPT-4, GPT-4o, GPT-4o-mini). Moreover, JSP achieves a state-of-the-art attack success rate of 92% on GPT-4 on the harmful query benchmark, and exhibits strong resistant to defence strategies. Warning: this paper contains offensive examples.

While speculative decoding has recently appeared as a promising direction for accelerating the inference of large language models (LLMs), the speedup and scalability are strongly bounded by the token acceptance rate. Prevalent methods usually organize predicted tokens as independent chains or fixed token trees, which fails to generalize to diverse query distributions. In this paper, we propose DySpec, a faster speculative decoding algorithm with a novel dynamic token tree structure. We begin by bridging the draft distribution and acceptance rate from intuitive and empirical clues, and successfully show that the two variables are strongly correlated. Based on this, we employ a greedy strategy to dynamically expand the token tree at run time. Theoretically, we show that our method can achieve optimal results under mild assumptions. Empirically, DySpec yields a higher acceptance rate and speedup than fixed trees. DySpec can drastically improve the throughput and reduce the latency of token generation across various data distribution and model sizes, which significantly outperforms strong competitors, including Specinfer and Sequoia. Under low temperature setting, DySpec can improve the throughput up to 9.1$\times$ and reduce the latency up to 9.4$\times$ on Llama2-70B. Under high temperature setting, DySpec can also improve the throughput up to 6.21$\times$, despite the increasing difficulty of speculating more than one token per step for draft model.

This paper presents the results of the LegalLens Shared Task, focusing on detecting legal violations within text in the wild across two sub-tasks: LegalLens-NER for identifying legal violation entities and LegalLens-NLI for associating these violations with relevant legal contexts and affected individuals. Using an enhanced LegalLens dataset covering labor, privacy, and consumer protection domains, 38 teams participated in the task. Our analysis reveals that while a mix of approaches was used, the top-performing teams in both tasks consistently relied on fine-tuning pre-trained language models, outperforming legal-specific models and few-shot methods. The top-performing team achieved a 7.11% improvement in NER over the baseline, while NLI saw a more marginal improvement of 5.7%. Despite these gains, the complexity of legal texts leaves room for further advancements.