We present LawBench, the first evaluation benchmark composed of 20 tasks aimed to assess the ability of Large Language Models (LLMs) to perform Chinese legal-related tasks. LawBench is meticulously crafted to enable precise assessment of LLMs’ legal capabilities from three cognitive levels that correspond to the widely accepted Bloom’s cognitive taxonomy. Using LawBench, we present a comprehensive evaluation of 21 popular LLMs and the first comparative analysis of the empirical results in order to reveal their relative strengths and weaknesses. All data, model predictions and evaluation code are accessible from https://github.com/open-compass/LawBench.

Large Language Models (LLMs) possess the potential to exert substantial influence on public perceptions and interactions with information. This raises concerns about the societal impact that could arise if the ideologies within these models can be easily manipulated. In this work, we investigate how effectively LLMs can learn and generalize ideological biases from their instruction-tuning data. Our findings reveal a concerning vulnerability: exposure to only a small amount of ideologically driven samples significantly alters the ideology of LLMs. Notably, LLMs demonstrate a startling ability to absorb ideology from one topic and generalize it to even unrelated ones. The ease with which LLMs’ ideologies can be skewed underscores the risks associated with intentionally poisoned training data by malicious actors or inadvertently introduced biases by data annotators. It also emphasizes the imperative for robust safeguards to mitigate the influence of ideological manipulations on LLMs.

In the realm of modern Large Language Models (LLMs), facilitating high-quality, multi-turn dialogues with humans represents a cornerstone feature. However, human-based evaluation of such a capability involves substantial manual effort. This study offers a formative assessment of current LLMs’ proficiency in emulating human-like, multi-turn conversations using an LLM-centric approach. The evaluation encompasses three key elements in the evaluation pipeline: utterance generation, evaluation protocol, and judgement, and we delve deeply into each aspect. GPT-4, both as an utterance generator and as a judge, exhibits exceptional performance. As a generator, GPT-4 crafts dialogues indistinguishable from human interactions in terms of style and flow. When judging, it shows a heightened alignment with human evaluative standards and consistency. Conversely, other LLMs face challenges in producing quality multi-turn dialogues, hindered by inadequate instruction-following abilities, a propensity for prolix utterances, and overall limited capabilities. Notably, generating extensive dialogues (e.g., spanning tens of turns) remains a formidable task for most LLMs, particularly in Chinese contexts. We hope that our work can serve as a valuable resource for evaluating the multi-turn chatting capabilities of LLMs. Related resources are available at https://github.com/open-compass/BotChat.

Recent advancements in large language models (LLMs) have showcased significant improvements in mathematics. However, traditional math benchmarks like GSM8k offer a unidimensional perspective, which fall short in providing a holistic assessment of the LLMs’ math capabilities. To address this gap, we introduce MathBench, a new benchmark that rigorously assesses the mathematical capabilities of large language models. MathBench spans a wide range of mathematical disciplines, offering a detailed evaluation of both theoretical understanding and practical problem-solving skills. The benchmark progresses through five distinct stages, from basic arithmetic to college mathematics, and is structured to evaluate models at various depths of knowledge. Each stage includes theoretical questions and application problems, allowing us to measure a model’s mathematical proficiency and its ability to apply concepts in practical scenarios. MathBench aims to enhance the evaluation of LLMs’ mathematical abilities, providing a nuanced view of their knowledge understanding levels and problem solving skills in a bilingual context.

We introduces ***LLaST***, a framework for building high-performance Large Language model based Speech-to-text Translation systems. We address the limitations of end-to-end speech translation (E2E ST) models by exploring model architecture design and optimization techniques tailored for LLMs. Our approach includes LLM-based speech translation architecture design, ASR-augmented training, multilingual data augmentation, and dual-LoRA optimization. Our approach demonstrates superior performance on the CoVoST-2 benchmark and showcases exceptional scaling capabilities powered by LLMs.We believe this effective method will serve as a strong baseline for speech translation and provide insights for futureimprovements of the LLM-based speech translation framework.

The large-scale conversational recommendation dataset is pivotal for the development of conversational recommender systems (CRS). Most existing CRS datasets suffers from the problems of data inextensibility and semantic inconsistency. To tackle these limitations and establish a benchmark in the conversational recommendation scenario, in this paper, we introduce the LLM-REDIAL dataset to facilitate the research in CRS. LLM-REDIAL is constructed by leveraging large language models (LLMs) to generate the high-quality dialogues. To provide the LLMs with detailed guidance, we integrate historical user behavior data with dialogue templates that are carefully designed through the combination of multiple pre-defined goals. LLM-REDIAL has two main advantages. First, it is the largest multi-domain CRS dataset which consists of 47.6k multi-turn dialogues with 482.6k utterances across 4 domains. Second, dialogue semantics and the users’ historical interaction information is highly consistent. Human evaluation are conducted to verify the quality of LLM-REDIAL. In addition, we evaluate the usability of advanced LLM-based models on LLM-REDIAL.

Open-sourced Large Language Models (LLMs) have achieved great success in various NLP tasks, however, they are still far inferior to API-based models when acting as agents. How to integrate agent ability into general LLMs becomes a crucial and urgent problem.This paper first delivers three key observations: (1) the current agent training corpus is entangled with both formats following and agent reasoning, which significantly shifts from the distribution of its pre-training data; (2) LLMs exhibit different learning speeds on the capabilities required by agent tasks; and (3) current approaches have side-effects when improving agent abilities by introducing hallucinations. Based on the above findings, we propose Agent-FLAN to effectively Fine-tune LANguage models for Agents.Through careful decomposition and redesign of the training corpus, Agent-FLAN enables Llama2-7B to outperform prior best works by 3.5% across various agent evaluation datasets. With comprehensively constructed negative samples, Agent-FLAN greatly alleviates the hallucination issues based on our established evaluation benchmark. Besides, it consistently improves the agent capability of LLMs when scaling model sizes while slightly enhancing the general capability of LLMs. The code and models are available at https://github.com/InternLM/Agent-FLAN.

Large Language Models (LLMs) have revolutionized various domains with extensive knowledge and creative capabilities. However, a critical issue with LLMs is their tendency to produce outputs that diverge from factual reality. This phenomenon is particularly concerning in sensitive applications such as medical consultation and legal advice, where accuracy is paramount. Inspired by human lie detectors using physiological responses, we introduce the LLM Factoscope, a novel Siamese network-based model that leverages the inner states of LLMs for factual detection. Our investigation reveals distinguishable patterns in LLMs’ inner states when generating factual versus non-factual content. We demonstrate its effectiveness across various architectures, achieving over 96% accuracy on our custom-collected factual detection dataset. Our work opens a new avenue for utilizing LLMs’ inner states for factual detection and encourages further exploration into LLMs’ inner workings for enhanced reliability and transparency.

Large language models (LLMs) have demonstrated impressive capabilities across various tasks, but their performance is highly sensitive to the prompts utilized. This variability poses challenges for accurate assessment and user satisfaction. Current research frequently overlooks instance-level prompt variations and their implications on subjective evaluations. To address these shortcomings, we introduce ProSA, a framework designed to evaluate and comprehend prompt sensitivity in LLMs. ProSA incorporates a novel sensitivity metric, PromptSensiScore, and leverages decoding confidence to elucidate underlying mechanisms. Our extensive study, spanning multiple tasks, uncovers that prompt sensitivity fluctuates across datasets and models, with larger models exhibiting enhanced robustness. We observe that few-shot examples can alleviate this sensitivity issue, and subjective evaluations are also susceptible to prompt sensitivities, particularly in complex, reasoning-oriented tasks. Furthermore, our findings indicate that higher model confidence correlates with increased prompt robustness. We believe this work will serve as a helpful tool in studying prompt sensitivity of LLMs. The project is released at: https://github.com/open-compass/ProSA.

Recently, the large language model (LLM) community has shown increasing interest in enhancing LLMs’ capability to handle extremely long documents. As various long-text techniques and model architectures emerge, the precise and detailed evaluation of models’ long-text capabilities has become increasingly important. Existing long-text evaluation benchmarks, such as L-Eval and LongBench, construct long-text test sets based on open-source datasets, focusing mainly on QA and summarization tasks. These datasets include test samples of varying lengths (from 2k to 32k+) entangled together, making it challenging to assess model capabilities across different length ranges. Moreover, they do not cover the ultralong settings (100k+ tokens) that the latest LLMs claim to achieve. In this paper, we introduce Ada-LEval, a length-adaptable benchmark for evaluating the long-context understanding of LLMs. Ada-LEval includes two challenging subsets, TSort and BestAnswer, which enable a more reliable evaluation of LLMs’ long context capabilities. These benchmarks support intricate manipulation of the length of test cases, and can easily produce text samples up to 128k tokens. We evaluate 4 state-of-the-art closed-source API models and 6 open-source models with Ada-LEval. The evaluation results demonstrate the limitations of current LLMs, especially in ultra-long-context settings. Our code is available at https://github.com/open-compass/Ada-LEval.

Reinforcement learning from human feedback (RLHF) is a vital strategy for enhancing model capability in language models. However, annotating preference data for RLHF is a resource-intensive and creativity-demanding process, while existing automatic generation methods face limitations in data diversity and quality. In response, we present Safer-Instruct, a novel pipeline for automatically constructing large-scale preference data. Our approach leverages reversed instruction tuning, instruction induction, and expert model evaluation to efficiently generate high-quality preference data without human annotators. To verify the effectiveness of Safer-Instruct, we apply the pipeline to construct a safety preference dataset as a case study. Finetuning an Alpaca model on this synthetic dataset not only demonstrates improved harmlessness but also outperforms models fine-tuned on human-annotated safety preference data, all the while maintaining a competitive edge in downstream tasks. Importantly, our Safer-Instruct framework is versatile and can be applied to generate preference data across various domains, extending its utility beyond safety preferences. It addresses the challenges in preference data acquisition and advances the development of more capable and responsible AI systems. For dataset and code implementation, see https://github.com/uscnlp-lime/safer-instruct/.

Temporal knowledge graph (TKG) reasoning has two settings: interpolation reasoning and extrapolation reasoning. Both of them draw plenty of research interest and have great significance. Methods of the former de-emphasize the temporal correlations among facts sequences, while methods of the latter require strict chronological order of knowledge and ignore inferring clues provided by missing facts of the past. These limit the practicability of TKG applications as almost all of the existing TKG reasoning methods are designed specifically to address either one setting. To this end, this paper proposes an original Temporal PAth-based Reasoning (TPAR) model for both the interpolation and extrapolation reasoning settings. TPAR performs a neural-driven symbolic reasoning fashion that is robust to ambiguous and noisy temporal data, and with fine interpretability as well. Comprehensive experiments show that TPAR outperforms SOTA methods on the link prediction task for both the interpolation and the extrapolation settings. A novel pipeline experimental setting is designed to evaluate the performances of SOTA combinations and the proposed TPAR towards interpolation and extrapolation reasoning. And more diverse experiments are conducted to show the robustness and interpretability of TPAR.

Reducing the ‘hallucination' problem of Large Language Models (LLMs) is crucial for their wide applications. A comprehensive and fine-grained measurement of the hallucination is the first key step for the governance of this issue but is under-explored in the community.Thus, we present ANAH, a bilingual dataset that offers ANalytical Annotation of Hallucinations in LLMs within Generative Question Answering.Each answer sentence in our dataset undergoes rigorous annotation, involving the retrieval of a reference fragment, the judgment of the hallucination type, and the correction of hallucinated content. ANAH consists of ~12k sentence-level annotations for ~4.3k LLM responses covering over 700 topics, constructed by a human-in-the-loop pipeline.Thanks to the fine granularity of the hallucination annotations, we can quantitatively confirm that the hallucinations of LLMs progressively accumulate in the answer and use ANAH to train and evaluate hallucination annotators. We conduct extensive experiments on studying generative and discriminative annotators and show that, although current open-source LLMs have difficulties in fine-grained hallucination annotation, the generative annotator trained with ANAH can surpass all open-source LLMs and GPT-3.5, obtain performance competitive with GPT-4, and exhibits better generalization ability on unseen questions.

Large language models (LLMs) have achieved remarkable performance on various NLP tasks and are augmented by tools for broader applications. Yet, how to evaluate and analyze the tool utilization capability of LLMs is still under-explored. In contrast to previous works that evaluate models holistically, we comprehensively decompose the tool utilization into multiple sub-processes, including instruction following, planning, reasoning, retrieval, understanding, and review. Based on that, we further introduce T-Eval to evaluate the tool-utilization capability step by step. T-Eval disentangles the tool utilization evaluation into several sub-domains along model capabilities, facilitating the inner understanding of both holistic and isolated competency of LLMs. We conduct extensive experiments on T-Eval and in-depth analysis of various LLMs. T-Eval not only exhibits consistency with the outcome-oriented evaluation but also provides a more fine-grained analysis of the capabilities of LLMs, providing a new perspective in LLM evaluation on tool-utilization ability. The benchmark will be available.

In recent years, Large Language Models (LLMs) have demonstrated exceptional performance in code-generation tasks. However, under enterprise scenarios where private APIs are pre-built, general LLMs often fail to meet expectations. Existing approaches are confronted with drawbacks of high resource consumption and inadequate handling of multi-API tasks. To address these challenges, we propose EpiGEN, an Efficient multi-Api code GENeration framework under enterprise scenario. It consists of three core modules: Task Decomposition Module (TDM), API Retrieval Module (ARM), and Code Generation Module (CGM), in which Langchain played an important role. Through a series of experiments, EpiGEN shows good acceptability and readability, compared to fully fine-tuned LLM with a larger number of parameters. Particularly, in medium and hard level tasks, the performance of EpiGEN on a single-GPU machine even surpasses that of a fully fine-tuned LLM that requires multi-GPU configuration. Generally, EpiGEN is model-size agnostic, facilitating a balance between the performance of code generation and computational requirements.

Unsupervised sentence representation learning is one of the fundamental problems in natural language processing with various downstream applications. Recently, contrastive learning has been widely adopted which derives high-quality sentence representations by pulling similar semantics closer and pushing dissimilar ones away. However, these methods fail to capture the fine-grained ranking information among the sentences, where each sentence is only treated as either positive or negative. In many real-world scenarios, one needs to distinguish and rank the sentences based on their similarities to a query sentence, e.g., very relevant, moderate relevant, less relevant, irrelevant, etc. In this paper, we propose a novel approach, RankCSE, for unsupervised sentence representation learning, which incorporates ranking consistency and ranking distillation with contrastive learning into a unified framework. In particular, we learn semantically discriminative sentence representations by simultaneously ensuring ranking consistency between two representations with different dropout masks, and distilling listwise ranking knowledge from the teacher. An extensive set of experiments are conducted on both semantic textual similarity (STS) and transfer (TR) tasks. Experimental results demonstrate the superior performance of our approach over several state-of-the-art baselines.

Temporal factors are tied to the growth of facts in realistic applications, such as the progress of diseases and the development of political situation, therefore, research on Temporal Knowledge Graph (TKG) attracks much attention. In TKG, relation patterns inherent with temporality are required to be studied for representation learning and reasoning across temporal facts. However, existing methods can hardly model temporal relation patterns, nor can capture the intrinsic connections between relations when evolving over time, lacking of interpretability. In this paper, we propose a novel temporal modeling method which represents temporal entities as Rotations in Quaternion Vector Space (RotateQVS) and relations as complex vectors in Hamilton’s quaternion space. We demonstrate our method can model key patterns of relations in TKG, such as symmetry, asymmetry, inverse, and can capture time-evolved relations by theory. And empirically, we show that our method can boost the performance of link prediction tasks over four temporal knowledge graph benchmarks.

Large-scale language models coupled with prompts have shown remarkable performance on few-shot learning. However, through systematic experiments, we find that the few-shot performance of small language models is poor, and using prompts on them brings fewer improvements than on larger ones. In this paper, we propose SMASH, an approach to improve SMAll language models’ few-SHot ability by training on intermediate tasks before prompt-based fine-tuning on downstream tasks. We design intermediate tasks for sentence-pair tasks and sentiment classification tasks by creating training examples with prompt templates similar to downstream tasks using sentences sampled from a large-scale unsupervised corpus, and apply knowledge distillation to distill from outputs of larger pre-trained models as the training objective. We conduct extensive experiments and show that SMASH can make a 6-layer DistilRoBRETa-base achieve comparable performance on few-shot datasets with a 12-layer RoBERTa-base at a low cost.

This paper describes novel models tailored for a new application, that of extracting the symptoms mentioned in clinical conversations along with their status. Lack of any publicly available corpus in this privacy-sensitive domain led us to develop our own corpus, consisting of about 3K conversations annotated by professional medical scribes. We propose two novel deep learning approaches to infer the symptom names and their status: (1) a new hierarchical span-attribute tagging (SA-T) model, trained using curriculum learning, and (2) a variant of sequence-to-sequence model which decodes the symptoms and their status from a few speaker turns within a sliding window over the conversation. This task stems from a realistic application of assisting medical providers in capturing symptoms mentioned by patients from their clinical conversations. To reflect this application, we define multiple metrics. From inter-rater agreement, we find that the task is inherently difficult. We conduct comprehensive evaluations on several contrasting conditions and observe that the performance of the models range from an F-score of 0.5 to 0.8 depending on the condition. Our analysis not only reveals the inherent challenges of the task, but also provides useful directions to improve the models.