UNComp: Uncertainty-Aware Long-Context Compressor for Efficient Large Language Model Inference

Deploying large language models (LLMs) is challenging due to their high memory and computational demands, especially during long-context inference. While key-value (KV) caching accelerates inference by reusing previously computed keys and values, it also introduces significant memory overhead. Existing KV cache compression methods such as eviction and merging typically compress the KV cache after it is generated and overlook the eviction of hidden states, failing to improve the speed of the prefilling stage. Additionally, applying a uniform compression rate across different attention heads can harm crucial retrieval heads in needle-in-a-haystack tasks due to excessive compression. In this paper, we propose UNComp, an uncertainty-aware compression scheme that leverages matrix entropy to estimate model uncertainty across layers and heads at the token sequence level. By grouping layers and heads based on their uncertainty, UNComp adaptively compresses both the hidden states and the KV cache. Our method achieves a 1.6x speedup in the prefilling stage and reduces the KV cache to 4.74% of its original size, resulting in a 6.4x increase in throughput and a 1.4x speedup in inference with only a 1.41% performance loss. Remarkably, in needle-in-a-haystack tasks, UNComp outperforms the full-size KV cache even when compressed to 9.38% of its original size. Our approach offers an efficient, training-free Grouped-Query Attention paradigm that can be seamlessly integrated into existing KV cache schemes.

NAVERO: Unlocking Fine-Grained Semantics for Video-Language Compositionality

We study the capability of Video-Language (VidL) models in understanding compositions between objects, attributes, actions and their relations. Composition understanding becomes particularly challenging for video data since the compositional relations rapidly change over time in videos. We first build a benchmark named AARO to evaluate composition understanding related to actions on top of spatial concepts. The benchmark is constructed by generating negative texts with incorrect action descriptions for a given video and the model is expected to pair a positive text with its corresponding video. Furthermore, we propose a training method called NAVERO which utilizes video-text data augmented with negative texts to enhance composition understanding. We also develop a negative-augmented visual-language matching loss which is used explicitly to benefit from the generated negative text. We compare NAVERO with other state-of-the-art methods in terms of compositional understanding as well as video-text retrieval performance. NAVERO achieves significant improvement over other methods for both video-language and image-language composition understanding, while maintaining strong performance on traditional text-video retrieval tasks.

Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabularies

Research on scaling large language models (LLMs) has primarily focused on model parameters and training data size, overlooking the role of vocabulary size. % Intuitively, larger vocabularies enable more efficient tokenization by representing sentences with fewer tokens, but they also increase the risk of under-fitting representations for rare tokens. We investigate how vocabulary size impacts LLM scaling laws by training models ranging from 33M to 3B parameters on up to 500B characters with various vocabulary configurations. We propose three complementary approaches for predicting the compute-optimal vocabulary size: IsoFLOPs analysis, derivative estimation, and parametric fit of the loss function. Our approaches converge on the same result that the optimal vocabulary size depends on the available compute budget and that larger models deserve larger vocabularies. However, most LLMs use too small vocabulary sizes. For example, we predict that the optimal vocabulary size of Llama2-70B should have been at least 216K, 7 times larger than its vocabulary of 32K. We validate our predictions empirically by training models with 3B parameters across different FLOPs budgets. Adopting our predicted optimal vocabulary size consistently improves downstream performance over commonly used vocabulary sizes. By increasing the vocabulary size from the conventional 32K to 43K, we improve performance on ARC-Challenge from 29.1 to 32.0 with the same 2.3e21 FLOPs. Our work emphasizes the necessity of jointly considering model parameters and vocabulary size for efficient scaling.

D2O:Dynamic Discriminative Operations for Efficient Generative Inference of Large Language Models

Efficient inference in Large Language Models (LLMs) is impeded by the growing memory demands of key-value (KV) caching, especially for longer sequences. Traditional KV cache eviction strategies, which prioritize less critical KV-pairs based on attention scores, often degrade generation quality, leading to issues such as context loss or hallucinations. To address this, we introduce Dynamic Discriminative Operations (D2O), a novel method that utilizes two-level discriminative strategies to optimize KV cache size without fine-tuning, while preserving essential context. Initially, by observing varying densities of attention weights between shallow and deep layers, we use this insight to determine which layers should avoid excessive eviction to minimize information loss. Subsequently, for the eviction strategy in each layer, D2O innovatively incorporates a compensation mechanism that maintains a similarity threshold to re-discriminate the importance of previously discarded tokens, determining whether they should be recalled and merged with similar tokens. Our approach not only achieves significant memory savings and enhances inference throughput by more than 3x but also maintains high-quality long-text generation. Extensive experiments across various benchmarks and LLM architectures have demonstrated that D2O significantly enhances performance with a constrained KV cache budget.

Rethinking Kullback-Leibler Divergence in Knowledge Distillation for Large Language Models

Kullback-Leiber divergence has been widely used in Knowledge Distillation (KD) to compress Large Language Models (LLMs). Contrary to prior assertions that reverse Kullback-Leibler (RKL) divergence is mode-seeking and thus preferable over the mean-seeking forward Kullback-Leibler (FKL) divergence, this study empirically and theoretically demonstrates that neither mode-seeking nor mean-seeking properties manifest in KD for LLMs. Instead, RKL and FKL are found to share the same optimization objective and both converge after a sufficient number of epochs. However, due to practical constraints, LLMs are seldom trained for such an extensive number of epochs. Meanwhile, we further find that RKL focuses on the tail part of the distributions, while FKL focuses on the head part at the beginning epochs. Consequently, we propose a simple yet effective Adaptive Kullback-Leiber (AKL) divergence method, which adaptively allocates weights to combine FKL and RKL. Metric-based and GPT-4-based evaluations demonstrate that the proposed AKL outperforms the baselines across various tasks and improves the diversity and quality of generated responses.

Electrocardiogram Instruction Tuning for Report Generation

Electrocardiogram (ECG) serves as the primary non-invasive diagnostic tool for cardiac conditions monitoring, are crucial in assisting clinicians. Recent studies have concentrated on classifying cardiac conditions using ECG data but have overlooked ECG report generation, which is not only time-consuming but also requires clinical expertise. To automate ECG report generation and ensure its versatility, we propose the Multimodal ECG Instruction Tuning (MEIT) framework, the \textit{first} attempt to tackle ECG report generation with LLMs and multimodal instructions. To facilitate future research, we establish a benchmark to evaluate MEIT with various LLMs backbones across two large-scale ECG datasets. Our approach uniquely aligns the representations of the ECG signal and the report, and we conduct extensive experiments to benchmark MEIT with nine open source LLMs, using more than 800,000 ECG reports. MEIT's results underscore the superior performance of instruction-tuned LLMs, showcasing their proficiency in quality report generation, zero-shot capabilities, and resilience to signal perturbation. These findings emphasize the efficacy of our MEIT framework and its potential for real-world clinical application.

RoboCodeX: Multimodal Code Generation for Robotic Behavior Synthesis

Robotic behavior synthesis, the problem of understanding multimodal inputs and generating precise physical control for robots, is an important part of Embodied AI. Despite successes in applying multimodal large language models for high-level understanding, it remains challenging to translate these conceptual understandings into detailed robotic actions while achieving generalization across various scenarios. In this paper, we propose a tree-structured multimodal code generation framework for generalized robotic behavior synthesis, termed RoboCodeX. RoboCodeX decomposes high-level human instructions into multiple object-centric manipulation units consisting of physical preferences such as affordance and safety constraints, and applies code generation to introduce generalization ability across various robotics platforms. To further enhance the capability to map conceptual and perceptual understanding into control commands, a specialized multimodal reasoning dataset is collected for pre-training and an iterative self-updating methodology is introduced for supervised fine-tuning. Extensive experiments demonstrate that RoboCodeX achieves state-of-the-art performance in both simulators and real robots on four different kinds of manipulation tasks and one navigation task.

A Spectral Perspective towards Understanding and Improving Adversarial Robustness

Deep neural networks (DNNs) are incredibly vulnerable to crafted, imperceptible adversarial perturbations. While adversarial training (AT) has proven to be an effective defense approach, the AT mechanism for robustness improvement is not fully understood. This work investigates AT from a spectral perspective, adding new insights to the design of effective defenses. In particular, we show that AT induces the deep model to focus more on the low-frequency region, which retains the shape-biased representations, to gain robustness. Further, we find that the spectrum of a white-box attack is primarily distributed in regions the model focuses on, and the perturbation attacks the spectral bands where the model is vulnerable. Based on this observation, to train a model tolerant to frequency-varying perturbation, we propose a spectral alignment regularization (SAR) such that the spectral output inferred by an attacked adversarial input stays as close as possible to its natural input counterpart. Experiments demonstrate that SAR and its weight averaging (WA) extension could significantly improve the robust accuracy by 1.14% ~ 3.87% relative to the standard AT, across multiple datasets (CIFAR-10, CIFAR-100 and Tiny ImageNet), and various attacks (PGD, C&W and Autoattack), without any extra data.

CrossGET: Cross-Guided Ensemble of Tokens for Accelerating Vision-Language Transformers

Vision-language models have achieved tremendous progress far beyond what we ever expected. However, their computational costs and latency are also dramatically growing with rapid development, making model acceleration exceedingly critical for researchers with limited resources and consumers with low-end devices. Although extensively studied for unimodal models, the acceleration for multimodal models, especially the vision-language Transformers, is still relatively under-explored. Accordingly, this paper proposes \textbf{Cross}-\textbf{G}uided \textbf{E}nsemble of \textbf{T}okens (\textbf{\emph{CrossGET}}) as a universal vison-language Transformer acceleration framework, which adaptively reduces token numbers during inference via cross-modal guidance on-the-fly, leading to significant model acceleration while keeping high performance. Specifically, the proposed \textit{CrossGET} has two key designs:1) \textit{Cross-Guided Matching and Ensemble}. \textit{CrossGET} incorporates cross-modal guided token matching and ensemble to merge tokens effectively, only introducing cross-modal tokens with negligible extra parameters. 2) \textit{Complete-Graph Soft Matching}. In contrast to the previous bipartite soft matching approach, \textit{CrossGET} introduces an efficient and effective complete-graph soft matching policy to achieve more reliable token-matching results. Extensive experiments on various vision-language tasks, datasets, and model architectures demonstrate the effectiveness and versatility of the proposed \textit{CrossGET} framework. The code will be at https://github.com/sdc17/CrossGET.

DyBit: Dynamic Bit-Precision Numbers for Efficient Quantized Neural Network Inference

To accelerate the inference of deep neural networks (DNNs), quantization with low-bitwidth numbers is actively researched. A prominent challenge is to quantize the DNN models into low-bitwidth numbers without significant accuracy degradation, especially at very low bitwidths (< 8 bits). This work targets an adaptive data representation with variable-length encoding called DyBit. DyBit can dynamically adjust the precision and range of separate bit-field to be adapted to the DNN weights/activations distribution. We also propose a hardware-aware quantization framework with a mixed-precision accelerator to trade-off the inference accuracy and speedup. Experimental results demonstrate that the inference accuracy via DyBit is 1.997% higher than the state-of-the-art at 4-bit quantization, and the proposed framework can achieve up to 8.1x speedup compared with the original model.