As LLMs make their way into many aspects of our lives, one place that warrants increased scrutiny with LLM usage is scientific research. Using LLMs for generating or analyzing data for research purposes is gaining popularity. But when such application is marred with ad-hoc decisions and engineering solutions, we need to be concerned about how it may affect that research, its findings, or any future works based on that research. We need a more scientific approach to using LLMs in our research. While there are several active efforts to support more systematic construction of prompts, they are often focused more on achieving desirable outcomes rather than producing replicable and generalizable knowledge with sufficient transparency, objectivity, or rigor. This article presents a new methodology inspired by codebook construction through qualitative methods to address that. Using humans in the loop and a multi-phase verification processes, this methodology lays a foundation for more systematic, objective, and trustworthy way of applying LLMs for analyzing data. Specifically, we show how a set of researchers can work through a rigorous process of labeling, deliberating, and documenting to remove subjectivity and bring transparency and replicability to prompt generation process.

In this paper, we introduce DiarizationLM, a framework to leverage large language models (LLM) to post-process the outputs from a speaker diarization system. Various goals can be achieved with the proposed framework, such as improving the readability of the diarized transcript, or reducing the word diarization error rate (WDER). In this framework, the outputs of the automatic speech recognition (ASR) and speaker diarization systems are represented as a compact textual format, which is included in the prompt to an optionally finetuned LLM. The outputs of the LLM can be used as the refined diarization results with the desired enhancement. As a post-processing step, this framework can be easily applied to any off-the-shelf ASR and speaker diarization systems without retraining existing components. Our experiments show that a finetuned PaLM 2-S model can reduce the WDER by rel. Speaker diarization is the task of partitioning speech into homogeneous segments according to speaker identities, answering the question "who spoken when" [1, 2]. Typical speaker diarization systems can be roughly categorized into two groups: modularized systems and end-to-end systems. A modularized speaker diarization system usually consists of multiple separately trained components including voice activity detection (VAD) [3, 4, 5, 6], speaker turn detection [7, 8], speaker encoder [9, 10, 11], and a clustering algorithm, which can be either unsupervised [12, 13, 14, 15, 16, 17] or supervised [18, 19]. In many real world applications such as meeting summarization, call center analysis, mobile recorder apps [24], and video captioning, knowing "who spoke when" is not sufficient. Speaker labels are more interpretable and meaningful when they are associated with speech transcripts.

As the scaling of Large Language Models (LLMs) has dramatically enhanced their capabilities, there has been a growing focus on the alignment problem to ensure their responsible and ethical use. While existing alignment efforts predominantly concentrate on universal values such as the HHH principle, the aspect of culture, which is inherently pluralistic and diverse, has not received adequate attention. This work introduces a new benchmark, CDEval, aimed at evaluating the cultural dimensions of LLMs. CDEval is constructed by incorporating both GPT-4's automated generation and human verification, covering six cultural dimensions across seven domains. Our comprehensive experiments provide intriguing insights into the culture of mainstream LLMs, highlighting both consistencies and variations across different dimensions and domains. The findings underscore the importance of integrating cultural considerations in LLM development, particularly for applications in diverse cultural settings. Through CDEval, we aim to broaden the horizon of LLM alignment research by including cultural dimensions, thus providing a more holistic framework for the future development and evaluation of LLMs. This benchmark serves as a valuable resource for cultural studies in LLMs, paving the way for more culturally aware and sensitive models.

Large Language Models (LLMs) are increasingly integrated into software applications. Downstream application developers often access LLMs through APIs provided as a service. However, LLM APIs are often updated silently and scheduled to be deprecated, forcing users to continuously adapt to evolving models. This can cause performance regression and affect prompt design choices, as evidenced by our case study on toxicity detection. Based on our case study, we emphasize the need for and re-examine the concept of regression testing for evolving LLM APIs. We argue that regression testing LLMs requires fundamental changes to traditional testing approaches, due to different correctness notions, prompting brittleness, and non-determinism in LLM APIs.

The quadratic time and memory complexity inherent to self-attention mechanisms, with respect to sequence length, presents a critical computational bottleneck in the training and deployment of large-scale Transformer-based language models. Recent theoretical results indicate the intractability of sub-quadratic softmax attention approximation under reasonable complexity assumptions. This paper addresses this challenge by first demonstrating that polynomial attention with high degree can effectively replace softmax without sacrificing model quality. Next, we develop polynomial sketching techniques from numerical linear algebra to achieve linear-time polynomial attention with approximation guarantees. Crucially, our approach achieves this speedup without requiring the sparsification of attention matrices. We also present a block-based algorithm to apply causal masking efficiently. Combining these techniques, we provide \emph{PolySketchFormer}, a practical linear-time Transformer architecture for language modeling that offers provable guarantees. We validate PolySketchFormer empirically by training language models capable of handling long contexts. These experiments utilize both synthetic and real-world datasets (PG19, Wikipedia and C4) on Google Cloud TPUs. For context lengths of 32k and GPT-2 style models, our model achieves a 2.5-4x speedup in training compared to FlashAttention, with no observed degradation in quality across our experiments.

We introduce AnyTool, a large language model agent designed to revolutionize the utilization of a vast array of tools in addressing user queries. We utilize over 16,000 APIs from Rapid API, operating under the assumption that a subset of these APIs could potentially resolve the queries. AnyTool primarily incorporates three elements: an API retriever with a hierarchical structure, a solver aimed at resolving user queries using a selected set of API candidates, and a self-reflection mechanism, which re-activates AnyTool if the initial solution proves impracticable. AnyTool is powered by the function calling feature of GPT-4, eliminating the need for training external modules. We also revisit the evaluation protocol introduced by previous works and identify a limitation in this protocol that leads to an artificially high pass rate. By revising the evaluation protocol to better reflect practical application scenarios, we introduce an additional benchmark, termed AnyToolBench. Experiments across various datasets demonstrate the superiority of our AnyTool over strong baselines such as ToolLLM and a GPT-4 variant tailored for tool utilization. For instance, AnyTool outperforms ToolLLM by +35.4% in terms of average pass rate on ToolBench. Code will be available at https://github.com/dyabel/AnyTool.

Automated red teaming holds substantial promise for uncovering and mitigating the risks associated with the malicious use of large language models (LLMs), yet the field lacks a standardized evaluation framework to rigorously assess new methods. To address this issue, we introduce HarmBench, a standardized evaluation framework for automated red teaming. We identify several desirable properties previously unaccounted for in red teaming evaluations and systematically design HarmBench to meet these criteria. Using HarmBench, we conduct a large-scale comparison of 18 red teaming methods and 33 target LLMs and defenses, yielding novel insights. We also introduce a highly efficient adversarial training method that greatly enhances LLM robustness across a wide range of attacks, demonstrating how HarmBench enables codevelopment of attacks and defenses. We open source HarmBench at https://github.com/centerforaisafety/HarmBench.

State-space models (SSMs), such as Mamba Gu & Dao (2034), have been proposed as alternatives to Transformer networks in language modeling, by incorporating gating, convolutions, and input-dependent token selection to mitigate the quadratic cost of multi-head attention. Although SSMs exhibit competitive performance, their in-context learning (ICL) capabilities, a remarkable emergent property of modern language models that enables task execution without parameter optimization, remain underexplored compared to Transformers. In this study, we evaluate the ICL performance of SSMs, focusing on Mamba, against Transformer models across various tasks. Our results show that SSMs perform comparably to Transformers in standard regression ICL tasks, while outperforming them in tasks like sparse parity learning. However, SSMs fall short in tasks involving non-standard retrieval functionality. To address these limitations, we introduce a hybrid model, \variant, that combines Mamba with attention blocks, surpassing individual models in tasks where they struggle independently. Our findings suggest that hybrid architectures offer promising avenues for enhancing ICL in language models.

The Transformer architecture has shown to be a powerful tool for a wide range of tasks. It is based on the self-attention mechanism, which is an inherently computationally expensive operation with quadratic computational complexity: memory usage and compute time increase quadratically with the length of the input sequences, thus limiting the application of Transformers. In this work, we propose a novel Clustering self-Attention mechanism using Surrogate Tokens (CAST), to optimize the attention computation and achieve efficient transformers. CAST utilizes learnable surrogate tokens to construct a cluster affinity matrix, used to cluster the input sequence and generate novel cluster summaries. The self-attention from within each cluster is then combined with the cluster summaries of other clusters, enabling information flow across the entire input sequence. CAST improves efficiency by reducing the complexity from $O(N^2)$ to $O(\alpha N)$ where N is the sequence length, and {\alpha} is constant according to the number of clusters and samples per cluster. We show that CAST performs better than or comparable to the baseline Transformers on long-range sequence modeling tasks, while also achieving higher results on time and memory efficiency than other efficient transformers.

Vision-Language Models (VLMs) have demonstrated their widespread viability thanks to extensive training in aligning visual instructions to answers. However, this conclusive alignment leads models to ignore critical visual reasoning, and further result in failures on meticulous visual problems and unfaithful responses. In this paper, we propose Chain of Manipulations, a mechanism that enables VLMs to solve problems with a series of manipulations, where each manipulation refers to an operation on the visual input, either from intrinsic abilities (e.g., grounding) acquired through prior training or from imitating human-like behaviors (e.g., zoom in). This mechanism encourages VLMs to generate faithful responses with evidential visual reasoning, and permits users to trace error causes in the interpretable paths. We thus train CogCoM, a general 17B VLM with a memory-based compatible architecture endowed this reasoning mechanism. Experiments show that our model achieves the state-of-the-art performance across 8 benchmarks from 3 categories, and a limited number of training steps with the data swiftly gains a competitive performance. The code and data are publicly available at https://github.com/THUDM/CogCoM.