Retrieval-Augmented Generation (RAG) can alleviate hallucinations of Large Language Models (LLMs) by referencing external documents. However, the misinformation in external documents may mislead LLMs' generation. To address this issue, we explore the task of "credibility-aware RAG", in which LLMs automatically adjust the influence of retrieved documents based on their credibility scores to counteract misinformation. To this end, we introduce a plug-and-play method named $\textbf{Cr}$edibility-aware $\textbf{A}$ttention $\textbf{M}$odification (CrAM). CrAM identifies influential attention heads in LLMs and adjusts their attention weights based on the credibility of the documents, thereby reducing the impact of low-credibility documents. Experiments on Natual Questions and TriviaQA using Llama2-13B, Llama3-8B, and Qwen-7B show that CrAM improves the RAG performance of LLMs against misinformation pollution by over 20%, even surpassing supervised fine-tuning methods.

Fine-tuning pre-trained Large Language Models (LLMs) is essential to align them with human values and intentions. This process often utilizes methods like pairwise comparisons and KL divergence against a reference LLM, focusing on the evaluation of full answers generated by the models. However, the generation of these responses occurs in a token level, following a sequential, auto-regressive fashion. In this paper, we introduce Token-level Direct Preference Optimization (TDPO), a novel approach to align LLMs with human preferences by optimizing policy at the token level. Unlike previous methods, which face challenges in divergence efficiency, TDPO incorporates forward KL divergence constraints for each token, improving alignment and diversity. Utilizing the Bradley-Terry model for a token-based reward system, TDPO enhances the regulation of KL divergence, while preserving simplicity without the need for explicit reward modeling. Experimental results across various text tasks demonstrate TDPO's superior performance in balancing alignment with generation diversity. Notably, fine-tuning with TDPO strikes a better balance than DPO in the controlled sentiment generation and single-turn dialogue datasets, and significantly improves the quality of generated responses compared to both DPO and PPO-based RLHF methods. Our code is open-sourced at https://github.com/Vance0124/Token-level-Direct-Preference-Optimization.

The BioNLP ACL'24 Shared Task on Streamlining Discharge Documentation aims to reduce the administrative burden on clinicians by automating the creation of critical sections of patient discharge letters. This paper presents our approach using the Llama3 8B quantized model to generate the "Brief Hospital Course" and "Discharge Instructions" sections. We employ a zero-shot method combined with Retrieval-Augmented Generation (RAG) to produce concise, contextually accurate summaries. Our contributions include the development of a curated template-based approach to ensure reliability and consistency, as well as the integration of RAG for word count prediction. We also describe several unsuccessful experiments to provide insights into our pathway for the competition. Our results demonstrate the effectiveness and efficiency of our approach, achieving high scores across multiple evaluation metrics.

Large Language Models (LLMs) have demonstrated impressive capability in many natural language tasks. However, the auto-regressive generation process makes LLMs prone to produce errors, hallucinations and inconsistent statements when performing multi-step reasoning. In this paper, by casting multi-step reasoning of LLMs as a heuristic search problem, we aim to alleviate the pathology by introducing Q*, a general, versatile and agile framework for guiding LLMs decoding process with deliberative planning. By learning a plug-and-play Q-value model as heuristic function for estimating expected future rewards, our Q* can effectively guide LLMs to select the most promising next reasoning step without fine-tuning LLMs for the current task, which avoids the significant computational overhead and potential risk of performance degeneration on other tasks. Extensive experiments on GSM8K, MATH and MBPP demonstrate the superiority of our method, contributing to improving the reasoning performance of existing open-source LLMs.

Visual programs are executable code generated by large language models to address visual reasoning problems. They decompose complex questions into multiple reasoning steps and invoke specialized models for each step to solve the problems. However, these programs are prone to logic errors, with our preliminary evaluation showing that 58% of the total errors are caused by program logic errors. Debugging complex visual programs remains a major bottleneck for visual reasoning. To address this, we introduce VDebugger, a novel critic-refiner framework trained to localize and debug visual programs by tracking execution step by step. VDebugger identifies and corrects program errors leveraging detailed execution feedback, improving interpretability and accuracy. The training data is generated through an automated pipeline that injects errors into correct visual programs using a novel mask-best decoding technique. Evaluations on six datasets demonstrate VDebugger's effectiveness, showing performance improvements of up to 3.2% in downstream task accuracy. Further studies show VDebugger's ability to generalize to unseen tasks, bringing a notable improvement of 2.3% on the unseen COVR task. Code, data and models are made publicly available at https://github.com/shirley-wu/vdebugger/

Large model training has been using recomputation to alleviate the memory pressure and pipelining to exploit the parallelism of data, tensor, and devices. The existing recomputation approaches may incur up to 40% overhead when training real-world models, e.g., the GPT model with 22B parameters. This is because they are executed on demand in the critical training path. In this paper, we design a new recomputation framework, Lynx, to reduce the overhead by overlapping the recomputation with communication occurring in training pipelines. It consists of an optimal scheduling algorithm (OPT) and a heuristic-based scheduling algorithm (HEU). OPT achieves a global optimum but suffers from a long search time. HEU was designed based on our observation that there are identical structures in large DNN models so that we can apply the same scheduling policy to all identical structures. HEU achieves a local optimum but reduces the search time by 99% compared to OPT. Our comprehensive evaluation using GPT models with 1.3B-20B parameters shows that both OPT and HEU outperform the state-of-the-art recomputation approaches (e.g., Megatron-LM and Checkmake) by 1.02-1.53x. HEU achieves a similar performance as OPT with a search time of 0.16s on average.

The recent proliferation of generative artificial intelligence (AI) technologies such as pre-trained large language models (LLMs) has opened up new frontiers in computational law. An exciting area of development is the use of AI to automate the deductive rule-based reasoning inherent in statutory and contract law. This paper argues that such automated deductive legal reasoning can now be viewed from the lens of software engineering, treating LLMs as interpreters of natural-language programs with natural-language inputs. We show how it is possible to apply principled software engineering techniques to enhance AI-driven legal reasoning of complex statutes and to unlock new applications in automated meta-reasoning such as mutation-guided example generation and metamorphic property-based testing.

It is likely that 2023 will be remembered as the year of Artificial Intelligence (AI). ChatGPT [2] was the fastest internet service to reach 100million users until now (May 2023) [3] and the technology of Large Language Models (LLMs) at its core is a fundamental element of sister apps for images such as Dall-e2, Midjourney and many others. One of the most fascinating aspects of LLMs is that they exhibit unpredicted emergent features. While the media excitedly reported how AI art generator have developed their own taste [4] or chatbots are able to pass school level exams in a growing number of disciplines [5], only in 2023 it was released that, for the past two years, GPT models had consistently improved its performance in tests designed to measure theory of mind in children [6]. For anyone familiar with complexity science, observing emergent properties in a complex system made of billions of artificial neurons is perhaps not surprising, but the growth in human-, or even superhuman-, like capabilities has attracted huge attention from the media and the public, sparking a hectic debate between the technology apocalyptic and integrated [7]. While it is clear that AI could bring us spectacular benefits, from better medical diagnosing to drug discovering, the risks have so far catalysed most of the public attention. Perils associated to narrow AI include increasing opportunities for manipulation of people, enhancing and dehumanising weapons, and rendering human labour increasingly obsolescent [8]. On the other hand, selfimproving "artificial general intelligence" (AGI) could pose an existential threat to humanity itself.

Large language models exhibit high-level commonsense reasoning abilities, especially with enhancement methods like Chain-of-Thought (CoT). However, we find these CoT-like methods lead to a considerable number of originally correct answers turning wrong, which we define as the Toxic CoT problem. To interpret and mitigate this problem, we first utilize attribution tracing and causal tracing methods to probe the internal working mechanism of the LLM during CoT reasoning. Through comparisons, we prove that the model exhibits information loss from the question over the shallow attention layers when generating rationales or answers. Based on the probing findings, we design a novel method called RIDERS (Residual decodIng and sERial-position Swap), which compensates for the information deficit in the model from both decoding and serial-position perspectives. Through extensive experiments on multiple commonsense reasoning benchmarks, we validate that this method not only significantly eliminates Toxic CoT problems (decreased by 23.6%), but also effectively improves the model's overall commonsense reasoning performance (increased by 5.5%).

In this work, we present empirical results regarding the feasibility of using offline large language models (LLMs) in the context of electronic design automation (EDA). The goal is to investigate and evaluate a contemporary language model's (Llama-2-7B) ability to function as a microelectronic Q & A expert as well as its reasoning, and generation capabilities in solving microelectronic-related problems. Llama-2-7B was tested across a variety of adaptation methods, including introducing a novel low-rank knowledge distillation (LoRA-KD) scheme. Our experiments produce both qualitative and quantitative results.