Large language models (LLMs) encapsulate a vast amount of factual information within their pre-trained weights, as evidenced by their ability to answer diverse questions across different domains. However, this knowledge is inherently limited, relying heavily on the characteristics of the training data. Consequently, using external datasets to incorporate new information or refine the capabilities of LLMs on previously seen information poses a significant challenge. In this study, we compare two common approaches: fine-tuning and retrieval-augmented generation (RAG). We evaluate both approaches on a variety of knowledge-intensive tasks across different topics. Our findings reveal that while fine-tuning offers some improvement, RAG consistently outperforms it, both for existing knowledge encountered during training and entirely new knowledge. Moreover, we find that LLMs struggle to learn new factual information through fine-tuning, and that exposing them to numerous variations of the same fact during training could alleviate this problem.
相關內容
《計算機信息》雜志發表高質量的論文，擴大了運籌學和計算的范圍，尋求有關理論、方法、實驗、系統和應用方面的原創研究論文、新穎的調查和教程論文，以及描述新的和有用的軟件工具的論文。官網鏈接： ·
大語言模型
·
MoDELS
·
motivation
·
樣例
·
Large language models (LLMs) are proficient at generating fluent text with minimal task-specific supervision. Yet, their ability to provide well-grounded rationalizations for knowledge-intensive tasks remains under-explored. Such tasks, like commonsense multiple-choice questions, require rationales based on world knowledge to support predictions and refute alternate options. We consider the task of generating knowledge-guided rationalization in natural language by using expert-written examples in a few-shot manner. Surprisingly, crowd-workers preferred knowledge-grounded rationales over crowdsourced rationalizations, citing their factuality, sufficiency, and comprehensive refutations. Although LLMs-generated rationales were preferable, further improvements in conciseness and novelty are required. In another study, we show how rationalization of incorrect model predictions erodes humans' trust in LLM-generated rationales. Motivated by these observations, we create a two-stage pipeline to review task predictions and eliminate potential incorrect decisions before rationalization, enabling trustworthy rationale generation.
Large language models (LLMs), especially Generative Pretrained Transformer (GPT) models, have significantly advanced in the industry in recent years. However, these models' broader development faces considerable challenges due to high operational and deployment costs. This has led to active research in improving the hardware efficiency of LLMs. Yet, the characteristics of real-world LLM workloads are often overlooked in current optimizations of LLM serving systems. In this work, we find that the absence of reliable workload data for evaluating LLM serving systems impacts the quality of service (QoS) and reliability in industrial deployments. This paper introduces the first real-world trace dataset of LLM serving workloads, detailing user, system, and LLM behaviors. We analyze this trace, highlighting burstiness, request and response distributions, and focusing on the reliability of GPT services. Based on this, we have developed a benchmark suite that reflects our dataset's workload patterns, enabling performance evaluation of serving systems. This suite captures the core patterns of workload distributions, allowing for precise scaling of the workload dataset to match system sizes. Our evaluation uncovers a previously unrecognized vulnerability of LLM serving systems to short-term burstiness, particularly in common workload scenarios. We observe that GPU memory limitations, caused by the fluctuating nature of burstiness, lead to significant performance degradation in existing LLM serving systems. Beyond benchmarking, understanding these patterns is valuable for optimizing LLM workload management, enabling elastic hardware resource adjustments to varying workloads. We will make the dataset and benchmark suite publicly available to encourage further research.
Large language models are meticulously aligned to be both helpful and harmless. However, recent research points to a potential overkill which means models may refuse to answer benign queries. In this paper, we investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to an over-attention of harmful words like 'kill' and prompts emphasizing safety will exacerbate overkill. Based on these insights, we introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon. We first extract such over-attention by amplifying the difference in the model's output distributions when responding to system prompts that either include or omit an emphasis on safety. Then we determine the final next-token predictions by downplaying the over-attention from the model via contrastive decoding. Empirical results indicate that our method has achieved an average reduction of the refusal rate by 20\% while having almost no impact on safety.
To achieve faithful reasoning that aligns with human expectations, large language models (LLMs) need to ground their reasoning to real-world knowledge (e.g., web facts, math and physical rules). Tools help LLMs access this external knowledge, but there remains challenges for fine-tuning LLM agents (e.g., Toolformer) to invoke tools in multi-step reasoning problems, where inter-connected tool calls require holistic and efficient tool usage planning. In this work, we propose a new method for LLMs to better leverage tools in multi-step reasoning. Our method, Chain-of-Abstraction (CoA), trains LLMs to first decode reasoning chains with abstract placeholders, and then call domain tools to reify each reasoning chain by filling in specific knowledge. This planning with abstract chains enables LLMs to learn more general reasoning strategies, which are robust to shifts of domain knowledge (e.g., math results) relevant to different reasoning questions. It also allows LLMs to perform decoding and calling of external tools in parallel, which avoids the inference delay caused by waiting for tool responses. In mathematical reasoning and Wiki QA domains, we show that our method consistently outperforms previous chain-of-thought and tool-augmented baselines on both in-distribution and out-of-distribution test sets, with an average ~6% absolute QA accuracy improvement. LLM agents trained with our method also show more efficient tool use, with inference speed being on average ~1.4x faster than baseline tool-augmented LLMs.
Large language models (LLM) are perceived to offer promising potentials for automating security tasks, such as those found in security operation centers (SOCs). As a first step towards evaluating this perceived potential, we investigate the use of LLMs in software pentesting, where the main task is to automatically identify software security vulnerabilities in source code. We hypothesize that an LLM-based AI agent can be improved over time for a specific security task as human operators interact with it. Such improvement can be made, as a first step, by engineering prompts fed to the LLM based on the responses produced, to include relevant contexts and structures so that the model provides more accurate results. Such engineering efforts become sustainable if the prompts that are engineered to produce better results on current tasks, also produce better results on future unknown tasks. To examine this hypothesis, we utilize the OWASP Benchmark Project 1.2 which contains 2,740 hand-crafted source code test cases containing various types of vulnerabilities. We divide the test cases into training and testing data, where we engineer the prompts based on the training data (only), and evaluate the final system on the testing data. We compare the AI agent's performance on the testing data against the performance of the agent without the prompt engineering. We also compare the AI agent's results against those from SonarQube, a widely used static code analyzer for security testing. We built and tested multiple versions of the AI agent using different off-the-shelf LLMs -- Google's Gemini-pro, as well as OpenAI's GPT-3.5-Turbo and GPT-4-Turbo (with both chat completion and assistant APIs). The results show that using LLMs is a viable approach to build an AI agent for software pentesting that can improve through repeated use and prompt engineering.
Although significant efforts have been dedicated to aligning large language models (LLMs), red-teaming reports suggest that these carefully aligned LLMs could still be jailbroken through adversarial prompts, tuning, or decoding. Upon examining the jailbreaking vulnerability of aligned LLMs, we observe that the decoding distributions of jailbroken and aligned models differ only in the initial generations. This observation motivates us to propose the weak-to-strong jailbreaking attack, where adversaries can utilize smaller unsafe/aligned LLMs (e.g., 7B) to guide jailbreaking against significantly larger aligned LLMs (e.g., 70B). To jailbreak, one only needs to additionally decode two smaller LLMs once, which involves minimal computation and latency compared to decoding the larger LLMs. The efficacy of this attack is demonstrated through experiments conducted on five models from three different organizations. Our study reveals a previously unnoticed yet efficient way of jailbreaking, exposing an urgent safety issue that needs to be considered when aligning LLMs. As an initial attempt, we propose a defense strategy to protect against such attacks, but creating more advanced defenses remains challenging. The code for replicating the method is available at //github.com/XuandongZhao/weak-to-strong
The reasoning abilities of large language models (LLMs) are the topic of a growing body of research in artificial intelligence and cognitive science. In this paper, we probe the extent to which a dozen LLMs are able to distinguish logically correct inferences from logically fallacious ones. We focus on inference patterns involving conditionals (e.g., 'If Ann has a queen, then Bob has a jack') and epistemic modals (e.g., 'Ann might have an ace', 'Bob must have a king'). These inference patterns have been of special interest to logicians, philosophers, and linguists, since they plausibly play a central role in human reasoning. Assessing LLMs on these inference patterns is thus highly relevant to the question of how much the reasoning abilities of LLMs match those of humans. Among the LLMs we tested, all but GPT-4 often make basic mistakes with conditionals. Moreover, even GPT-4 displays logically inconsistent judgments across inference patterns involving epistemic modals.
Large language models (LLMs) have exhibited an array of reasoning capabilities but face challenges like error propagation and hallucination, particularly in specialised areas like finance, where data is heterogeneous, and precision is paramount. We explore the potential of language model augmentation with external tools to mitigate these limitations and offload certain reasoning steps to external tools that are more suited for the task, instead of solely depending on the LLM's inherent abilities. More concretely, using financial domain question-answering datasets, we apply supervised fine-tuning on a LLaMA-2 13B Chat model to act both as a 'task router' and 'task solver'. The 'task router' dynamically directs a question to either be answered internally by the LLM or externally via the right tool from the tool set. Our tool-equipped SFT model, Raven, demonstrates an improvement of 35.2% and 5.06% over the base model and SFT-only baselines, respectively, and is highly competitive with strong GPT-3.5 results. To the best of our knowledge, our work is the first that investigates tool augmentation of language models for the finance domain.
Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this paper, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations, and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional machine learning models.
Pre-trained language representation models, such as BERT, capture a general language representation from large-scale corpora, but lack domain-specific knowledge. When reading a domain text, experts make inferences with relevant knowledge. For machines to achieve this capability, we propose a knowledge-enabled language representation model (K-BERT) with knowledge graphs (KGs), in which triples are injected into the sentences as domain knowledge. However, too much knowledge incorporation may divert the sentence from its correct meaning, which is called knowledge noise (KN) issue. To overcome KN, K-BERT introduces soft-position and visible matrix to limit the impact of knowledge. K-BERT can easily inject domain knowledge into the models by equipped with a KG without pre-training by-self because it is capable of loading model parameters from the pre-trained BERT. Our investigation reveals promising results in twelve NLP tasks. Especially in domain-specific tasks (including finance, law, and medicine), K-BERT significantly outperforms BERT, which demonstrates that K-BERT is an excellent choice for solving the knowledge-driven problems that require experts.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191