Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2x, most configurations can achieve competitive performance to and higher throughput than standard transformers, but when further reducing the size of the KV cache, pairing queries of all layers with KVs of upper layers can better maintain performance, although it also introduces additional training cost and prefilling latency. We hope that this work will help users choose the appropriate approach according to their requirements and facilitate research on the acceleration of LLM inference.

Language models are often trained to maximize the likelihood of the next token given past tokens in the training dataset. However, during inference time, they are utilized differently, generating text sequentially and auto-regressively by using previously generated tokens as input to predict the next one. Marginal differences in predictions at each step can cascade over successive steps, resulting in different distributions from what the models were trained for and potentially leading to unpredictable behavior. This paper proposes two simple approaches based on model own generation to address this discrepancy between the training and inference time. Our first approach is Batch-Scheduled Sampling, where, during training, we stochastically choose between the ground-truth token from the dataset and the model's own generated token as input to predict the next token. This is done in an offline manner, modifying the context window by interleaving ground-truth tokens with those generated by the model. Our second approach is Reference-Answer-based Correction, where we explicitly incorporate a self-correction capability into the model during training. This enables the model to effectively self-correct the gaps between the generated sequences and the ground truth data without relying on an external oracle model. By incorporating our proposed strategies during training, we have observed an overall improvement in performance compared to baseline methods, as demonstrated by our extensive experiments using summarization, general question-answering, and math question-answering tasks.

Existing automated essay scoring (AES) has solely relied on essay text without using explanatory rationales for the scores, thereby forgoing an opportunity to capture the specific aspects evaluated by rubric indicators in a fine-grained manner. This paper introduces Rationale-based Multiple Trait Scoring (RMTS), a novel approach for multi-trait essay scoring that integrates prompt-engineering-based large language models (LLMs) with a fine-tuning-based essay scoring model using a smaller large language model (S-LLM). RMTS uses an LLM-based trait-wise rationale generation system where a separate LLM agent generates trait-specific rationales based on rubric guidelines, which the scoring model uses to accurately predict multi-trait scores. Extensive experiments on benchmark datasets, including ASAP, ASAP++, and Feedback Prize, show that RMTS significantly outperforms state-of-the-art models and vanilla S-LLMs in trait-specific scoring. By assisting quantitative assessment with fine-grained qualitative rationales, RMTS enhances the trait-wise reliability, providing partial explanations about essays.

Ranking consistently emerges as a primary focus in information retrieval research. Retrieval and ranking models serve as the foundation for numerous applications, including web search, open domain QA, enterprise domain QA, and text-based recommender systems. Typically, these models undergo training on triplets consisting of binary relevance assignments, comprising one positive and one negative passage. However, their utilization involves a context where a significantly more nuanced understanding of relevance is necessary, especially when re-ranking a large pool of potentially relevant passages. Although collecting positive examples through user feedback like impressions or clicks is straightforward, identifying suitable negative pairs from a vast pool of possibly millions or even billions of documents possess a greater challenge. Generating a substantial number of negative pairs is often necessary to maintain the high quality of the model. Several approaches have been suggested in literature to tackle the issue of selecting suitable negative pairs from an extensive corpus. This study focuses on explaining the crucial role of hard negatives in the training process of cross-encoder models, specifically aiming to explain the performance gains observed with hard negative sampling compared to random sampling. We have developed a robust hard negative mining technique for efficient training of cross-encoder re-rank models on an enterprise dataset which has domain specific context. We provide a novel perspective to enhance retrieval models, ultimately influencing the performance of advanced LLM systems like Retrieval-Augmented Generation (RAG) and Reasoning and Action Agents (ReAct). The proposed approach demonstrates that learning both similarity and dissimilarity simultaneously with cross-encoders improves performance of retrieval systems.

Client-Service Representatives (CSRs) are vital to organizations. Frequent interactions with disgruntled clients, however, disrupt their mental well-being. To help CSRs regulate their emotions while interacting with uncivil clients, we designed Pro-Pilot, an LLM-powered assistant, and evaluated its efficacy, perception, and use. Our comparative analyses between 665 human and Pro-Pilot-generated support messages demonstrate Pro-Pilot's ability to adapt to and demonstrate empathy in various incivility incidents. Additionally, 143 CSRs assessed Pro-Pilot's empathy as more sincere and actionable than human messages. Finally, we interviewed 20 CSRs who interacted with Pro-Pilot in a simulation exercise. They reported that Pro-Pilot helped them avoid negative thinking, recenter thoughts, and humanize clients; showing potential for bridging gaps in coworker support. Yet, they also noted deployment challenges and emphasized the irreplaceability of shared experiences. We discuss future designs and societal implications of AI-mediated emotional labor, underscoring empathy as a critical function for AI assistants in front-office roles.

Long-context efficiency has recently become a trending topic in serving large language models (LLMs). And mixture of depths (MoD) is proposed as a perfect fit to bring down both latency and memory. In this paper, however, we discover that MoD can barely transform existing LLMs without costly training over an extensive number of tokens. To enable the transformations from any LLMs to MoD ones, we showcase top-k operator in MoD should be promoted to threshold-p operator, and refinement to architecture and data should also be crafted along. All these designs form our method termed MoDification. Through a comprehensive set of experiments covering model scales from 3B to 70B, we exhibit MoDification strikes an excellent balance between efficiency and effectiveness. MoDification can achieve up to ~1.2x speedup in latency and ~1.8x reduction in memory compared to original LLMs especially in long-context applications.

Medical task-oriented dialogue systems can assist doctors by collecting patient medical history, aiding in diagnosis, or guiding treatment selection, thereby reducing doctor burnout and expanding access to medical services. However, doctor-patient dialogue datasets are not readily available, primarily due to privacy regulations. Moreover, existing datasets lack comprehensive annotations involving medical slots and their different attributes, such as symptoms and their onset, progression, and severity. These comprehensive annotations are crucial for accurate diagnosis. Finally, most existing datasets are non-English, limiting their utility for the larger research community. In response, we introduce MediTOD, a new dataset of doctor-patient dialogues in English for the medical history-taking task. Collaborating with doctors, we devise a questionnaire-based labeling scheme tailored to the medical domain. Then, medical professionals create the dataset with high-quality comprehensive annotations, capturing medical slots and their attributes. We establish benchmarks in supervised and few-shot settings on MediTOD for natural language understanding, policy learning, and natural language generation subtasks, evaluating models from both TOD and biomedical domains. We make MediTOD publicly available for future research.

Perceiving and generating diverse modalities are crucial for AI models to effectively learn from and engage with real-world signals, necessitating reliable evaluations for their development. We identify two major issues in current evaluations: (1) inconsistent standards, shaped by different communities with varying protocols and maturity levels; and (2) significant query, grading, and generalization biases. To address these, we introduce MixEval-X, the first any-to-any, real-world benchmark designed to optimize and standardize evaluations across diverse input and output modalities. We propose multi-modal benchmark mixture and adaptation-rectification pipelines to reconstruct real-world task distributions, ensuring evaluations generalize effectively to real-world use cases. Extensive meta-evaluations show our approach effectively aligns benchmark samples with real-world task distributions. Meanwhile, MixEval-X's model rankings correlate strongly with that of crowd-sourced real-world evaluations (up to 0.98) while being much more efficient. We provide comprehensive leaderboards to rerank existing models and organizations and offer insights to enhance understanding of multi-modal evaluations and inform future research.

The development of 3D medical vision-language models holds significant potential for disease diagnosis and patient treatment. However, compared to 2D medical images, 3D medical images, such as CT scans, face challenges related to limited training data and high dimension, which severely restrict the progress of 3D medical vision-language models. To address these issues, we collect a large amount of unlabeled 3D CT data and utilize self-supervised learning to construct a 3D visual foundation model for extracting 3D visual features. Then, we apply 3D spatial convolutions to aggregate and project high-level image features, reducing computational complexity while preserving spatial information. We also construct two instruction-tuning datasets based on BIMCV-R and CT-RATE to fine-tune the 3D vision-language model. Our model demonstrates superior performance compared to existing methods in report generation, visual question answering, and disease diagnosis. Code and data will be made publicly available soon.

We introduce Controller-Embedded Language Model Interactions (CELI), a framework that integrates control logic directly within language model (LM) prompts, facilitating complex, multi-stage task execution. CELI addresses limitations of existing prompt engineering and workflow optimization techniques by embedding control logic directly within the operational context of language models, enabling dynamic adaptation to evolving task requirements. Our framework transfers control from the traditional programming execution environment to the LMs, allowing them to autonomously manage computational workflows while maintaining seamless interaction with external systems and functions. CELI supports arbitrary function calls with variable arguments, bridging the gap between LMs' adaptive reasoning capabilities and conventional software paradigms' structured control mechanisms. To evaluate CELI's versatility and effectiveness, we conducted case studies in two distinct domains: code generation (HumanEval benchmark) and multi-stage content generation (Wikipedia-style articles). The results demonstrate notable performance improvements across a range of domains. CELI achieved a 4.9 percentage point improvement over the best reported score of the baseline GPT-4 model on the HumanEval code generation benchmark. In multi-stage content generation, 94.4% of CELI-produced Wikipedia-style articles met or exceeded first draft quality when optimally configured, with 44.4% achieving high quality. These outcomes underscore CELI's potential for optimizing AI-driven workflows across diverse computational domains.