Large Language Models (LLMs) have been widely deployed in a variety of applications, and the context length is rapidly increasing to handle tasks such as long-document QA and complex logical reasoning. However, long context poses significant challenges for inference efficiency, including high memory costs of key-value (KV) cache and increased latency due to extensive memory accesses. Recent works have proposed compressing KV cache to approximate computation, but these methods either evict tokens permanently, never recalling them for later inference, or recall previous tokens at the granularity of pages divided by textual positions. Both approaches degrade the model accuracy and output quality. To achieve efficient and accurate recallable KV cache compression, we introduce ClusterKV, which recalls tokens at the granularity of semantic clusters. We design and implement efficient algorithms and systems for clustering, selection, indexing and caching. Experiment results show that ClusterKV attains negligible accuracy loss across various tasks with 32k context lengths, using only a 1k to 2k KV cache budget, and achieves up to a 2$\times$ speedup in latency and a 2.5$\times$ improvement in decoding throughput. Compared to SoTA recallable KV compression methods, ClusterKV demonstrates higher model accuracy and output quality, while maintaining or exceeding inference efficiency.

While fusing heterogeneous open-source LLMs with varying architectures and sizes can potentially integrate the strengths of different models, existing fusion methods face significant challenges, such as vocabulary alignment and merging distribution matrices. These procedures are not only complex but also prone to introducing noise and errors. In this paper, we propose an implicit fusion method, Weighted-Reward Preference Optimization (WRPO), which leverages preference optimization between the source LLMs and the target LLM to transfer their capabilities effectively. WRPO eliminates the need for vocabulary alignment and matrix fusion and can be efficiently scaled to accommodate various LLMs. To address distributional deviations between the source and target LLMs, WRPO introduces a progressive adaptation strategy that gradually shifts reliance on preferred examples from the target LLM to the source LLMs. Extensive experiments on the MT-Bench, AlpacaEval-2, and Arena-Hard benchmarks demonstrate that WRPO consistently outperforms existing knowledge fusion methods and various fine-tuning baselines. When applied to LLaMA3-8B-Instruct as the target model, WRPO achieves a length-controlled win rate of 55.9% against GPT-4-Preview-1106 on AlpacaEval-2 and a win rate of 46.2% against GPT-4-0314 on Arena-Hard. Our code is available at https://github.com/SLIT-AI/WRPO. Combining the strengths of multiple Large Language Models (LLMs) can potentially enhance the capabilities of individual models. Model ensemble techniques (Jiang et al., 2023b; Wang et al., 2024b) aggregate predictions from several models to improve overall performance and robustness over a single model. However, this approach requires substantial computational resources, as all models must remain active during inference. The Mixture of Experts (MoE) (Komatsuzaki et al., 2023; Feng et al., 2024; Sukhbaatar et al., 2024) leverages sparse expert networks to boost capacity by activating only a subset of parameters. Despite reduced activation, MoEs still incur significant memory overhead, as all parameters must be maintained. Model merging (Wortsman et al., 2022; Matena & Raffel, 2022; Yadav et al., 2024), which combines independently trained instances of the same model through arithmetic operations, allows a single model to be maintained during inference.

Uncertainty quantification in text-to-image (T2I) generative models is crucial for understanding model behavior and improving output reliability. In this paper, we are the first to quantify and evaluate the uncertainty of T2I models with respect to the prompt. Alongside adapting existing approaches designed to measure uncertainty in the image space, we also introduce Prompt-based UNCertainty Estimation for T2I models (PUNC), a novel method leveraging Large Vision-Language Models (LVLMs) to better address uncertainties arising from the semantics of the prompt and generated images. PUNC utilizes a LVLM to caption a generated image, and then compares the caption with the original prompt in the more semantically meaningful text space. PUNC also enables the disentanglement of both aleatoric and epistemic uncertainties via precision and recall, which image-space approaches are unable to do. Extensive experiments demonstrate that PUNC outperforms state-of-the-art uncertainty estimation techniques across various settings. Uncertainty quantification in text-to-image generation models can be used on various applications including bias detection, copyright protection, and OOD detection. We also introduce a comprehensive dataset of text prompts and generation pairs to foster further research in uncertainty quantification for generative models. Our findings illustrate that PUNC not only achieves competitive performance but also enables novel applications in evaluating and improving the trustworthiness of text-to-image models.

Tokenization is an important preprocessing step in the training and inference of large language models (LLMs). While there has been extensive research on the expressive power of the neural achitectures used in LLMs, the impact of tokenization has not been well understood. In this work, we demonstrate that tokenization, irrespective of the algorithm used, acts as an inverse homomorphism between strings and tokens. This suggests that the character space of the source language and the token space of the tokenized language are homomorphic, preserving the structural properties of the source language. Additionally, we explore the concept of proper tokenization, which refers to an unambiguous tokenization returned from the tokenizer. Our analysis reveals that the expressiveness of neural architectures in recognizing context-free languages is not affected by tokenization.

Artificial intelligence has, so far, largely automated routine tasks, but what does it mean for the future of work if Large Language Models (LLMs) show creativity comparable to humans? To measure the creativity of LLMs holistically, the current study uses 13 creative tasks spanning three domains. We benchmark the LLMs against individual humans, and also take a novel approach by comparing them to the collective creativity of groups of humans. We find that the best LLMs (Claude and GPT-4) rank in the 52nd percentile against humans, and overall LLMs excel in divergent thinking and problem solving but lag in creative writing. When questioned 10 times, an LLM's collective creativity is equivalent to 8-10 humans. When more responses are requested, two additional responses of LLMs equal one extra human. Ultimately, LLMs, when optimally applied, may compete with a small group of humans in the future of work.

Large language models (LLMs) have demonstrated impressive capabilities in role-playing tasks. However, there is limited research on whether LLMs can accurately simulate user behavior in real-world scenarios, such as social media. This requires models to effectively analyze a user's history and simulate their role. In this paper, we introduce \textbf{FineRob}, a novel fine-grained behavior simulation dataset. We collect the complete behavioral history of 1,866 distinct users across three social media platforms. Each behavior is decomposed into three fine-grained elements: object, type, and content, resulting in 78.6k QA records. Based on FineRob, we identify two dominant reasoning patterns in LLMs' behavior simulation processes and propose the \textbf{OM-CoT} fine-tuning method to enhance the capability. Through comprehensive experiments, we conduct an in-depth analysis of key factors of behavior simulation and also demonstrate the effectiveness of OM-CoT approach\footnote{Code and dataset are available at \url{https://github.com/linkseed18612254945/FineRob}}

Large language models (LLMs) demonstrate exceptional performance but incur high serving costs due to substantial memory demands, with the key-value (KV) cache being a primary bottleneck. Existing KV cache compression methods, including quantization and pruning, struggle with limitations such as uniform treatment of keys and values and static memory allocation across attention heads. To address these challenges, we introduce LeanKV, a unified KV cache compression framework that enhances LLM serving efficiency without compromising accuracy through three innovations: (1) Hetero-KV quantization, which stores keys at a higher precision than values to reflect their greater impact on attention computations; (2) per-head dynamic sparsity, which allocates memory based on token importance per head and per request; and (3) unified KV compression, integrating mixed-precision quantization and selective pruning to enable a smooth tradeoff between model accuracy and memory efficiency. To efficiently support these techniques, LeanKV introduces systems optimizations including unified paging and on-GPU parallel memory management. Implemented on vLLM, LeanKV compresses the KV cache by $3.0\times$ to $5.0\times$ without accuracy loss and up to $11.0\times$ with under 5% accuracy loss, enhancing throughput by $1.9\times$ to $2.5\times$, and up to $6.9\times$.

Large Language Models (LLMs) are widely used in complex natural language processing tasks but raise privacy and security concerns due to the lack of identity recognition. This paper proposes a multi-party credible watermarking framework (CredID) involving a trusted third party (TTP) and multiple LLM vendors to address these issues. In the watermark embedding stage, vendors request a seed from the TTP to generate watermarked text without sending the user's prompt. In the extraction stage, the TTP coordinates each vendor to extract and verify the watermark from the text. This provides a credible watermarking scheme while preserving vendor privacy. Furthermore, current watermarking algorithms struggle with text quality, information capacity, and robustness, making it challenging to meet the diverse identification needs of LLMs. Thus, we propose a novel multi-bit watermarking algorithm and an open-source toolkit to facilitate research. Experiments show our CredID enhances watermark credibility and efficiency without compromising text quality. Additionally, we successfully utilized this framework to achieve highly accurate identification among multiple LLM vendors.

Recent advancements in large language models (LLMs) have significantly enhanced the ability of LLM-based systems to perform complex tasks through natural language processing and tool interaction. However, optimizing these LLM-based systems for specific tasks remains challenging, often requiring manual interventions like prompt engineering and hyperparameter tuning. Existing automatic optimization methods, such as textual feedback-based techniques (e.g., TextGrad), tend to focus on immediate feedback, analogous to using immediate derivatives in traditional numerical gradient descent. However, relying solely on such feedback can be limited when the adjustments made in response to this feedback are either too small or fluctuate irregularly, potentially slowing down or even stalling the optimization process. To overcome these challenges, more adaptive methods are needed, especially in situations where the system's response is evolving slowly or unpredictably. In this paper, we introduce REVOLVE, an optimization method that tracks how "R"esponses "EVOLVE" across iterations in LLM systems. By focusing on the evolution of responses over time, REVOLVE enables more stable and effective optimization by making thoughtful, progressive adjustments at each step. Experimental results demonstrate that REVOLVE outperforms competitive baselines, achieving a 7.8% improvement in prompt optimization, a 20.72% gain in solution refinement, and a 29.17% increase in code optimization. Additionally, REVOLVE converges in fewer iterations, resulting in significant computational savings. These advantages highlight its adaptability and efficiency, positioning REVOLVE as a valuable tool for optimizing LLM-based systems and accelerating the development of next-generation AI technologies. Code is available at: https://github.com/Peiyance/REVOLVE.

We present TokenFlow, a novel unified image tokenizer that bridges the long-standing gap between multimodal understanding and generation. Prior research attempt to employ a single reconstruction-targeted Vector Quantization (VQ) encoder for unifying these two tasks. We observe that understanding and generation require fundamentally different granularities of visual information. This leads to a critical trade-off, particularly compromising performance in multimodal understanding tasks. TokenFlow addresses this challenge through an innovative dual-codebook architecture that decouples semantic and pixel-level feature learning while maintaining their alignment via a shared mapping mechanism. This design enables direct access to both high-level semantic representations crucial for understanding tasks and fine-grained visual features essential for generation through shared indices. Our extensive experiments demonstrate TokenFlow's superiority across multiple dimensions. Leveraging TokenFlow, we demonstrate for the first time that discrete visual input can surpass LLaVA-1.5 13B in understanding performance, achieving a 7.2\% average improvement. For image reconstruction, we achieve a strong FID score of 0.63 at 384*384 resolution. Moreover, TokenFlow establishes state-of-the-art performance in autoregressive image generation with a GenEval score of 0.55 at 256*256 resolution, achieving comparable results to SDXL.