Linge Li

Zunhai Su, Zhe Chen, Wang Shen et al.

Key-Value (KV) cache facilitates efficient large language models (LLMs) inference by avoiding recomputation of past KVs. As the batch size and context length increase, the oversized KV caches become a significant memory bottleneck, highlighting the need for efficient compression. Existing KV quantization rely on fine-grained quantization or the retention of a significant portion of high bit-widths caches, both of which compromise compression ratio and often fail to maintain robustness at extremely low average bit-widths. In this work, we explore the potential of rotation technique for 2-bit KV quantization and propose RotateKV, which achieves accurate and robust performance through the following innovations: (i) Outlier-Aware Rotation, which utilizes channel-reordering to adapt the rotations to varying channel-wise outlier distributions without sacrificing the computational efficiency of the fast Walsh-Hadamard transform (FWHT); (ii) Pre-RoPE Grouped-Head Rotation, which mitigates the impact of rotary position embedding (RoPE) on proposed outlier-aware rotation and further smooths outliers across heads; (iii) Attention-Sink-Aware Quantization, which leverages the massive activations to precisely identify and protect attention sinks. RotateKV achieves less than 0.3 perplexity (PPL) degradation with 2-bit quantization on WikiText-2 using LLaMA-2-13B, maintains strong CoT reasoning and long-context capabilities, with less than 1.7\% degradation on GSM8K, outperforming existing methods even at lower average bit-widths. RotateKV also showcases a 3.97x reduction in peak memory usage, supports 5.75x larger batch sizes, and achieves a 2.32x speedup in decoding stage.

Zunhai Su, Wang Shen, Linge Li et al.

Vision-language models (VLMs) show remarkable performance in multimodal tasks. However, excessively long multimodal inputs lead to oversized Key-Value (KV) caches, resulting in significant memory consumption and I/O bottlenecks. Previous KV quantization methods for Large Language Models (LLMs) may alleviate these issues but overlook the attention saliency differences of multimodal tokens, resulting in suboptimal performance. In this paper, we investigate the attention-aware token saliency patterns in VLM and propose AKVQ-VL. AKVQ-VL leverages the proposed Text-Salient Attention (TSA) and Pivot-Token-Salient Attention (PSA) patterns to adaptively allocate bit budgets. Moreover, achieving extremely low-bit quantization requires effectively addressing outliers in KV tensors. AKVQ-VL utilizes the Walsh-Hadamard transform (WHT) to construct outlier-free KV caches, thereby reducing quantization difficulty. Evaluations of 2-bit quantization on 12 long-context and multimodal tasks demonstrate that AKVQ-VL maintains or even improves accuracy, outperforming LLM-oriented methods. AKVQ-VL can reduce peak memory usage by 2.13x, support up to 3.25x larger batch sizes and 2.46x throughput.

Xingjian Tang, Jingwei Guan, Linge Li et al.

Diffusion models, as powerful generative models, have found a wide range of applications and shown great potential in solving image reconstruction problems. Some works attempted to solve MRI reconstruction with diffusion models, but these methods operate directly in pixel space, leading to higher computational costs for optimization and inference. Latent diffusion models, pre-trained on natural images with rich visual priors, are expected to solve the high computational cost problem in MRI reconstruction by operating in a lower-dimensional latent space. However, direct application to MRI reconstruction faces three key challenges: (1) absence of explicit control mechanisms for medical fidelity, (2) domain gap between natural images and MR physics, and (3) undefined data consistency in latent space. To address these challenges, a novel Latent Diffusion Prior-based undersampled MRI reconstruction (LDPM) method is proposed. Our LDPM framework addresses these challenges by: (1) a sketch-guided pipeline with a two-step reconstruction strategy, which balances perceptual quality and anatomical fidelity, (2) an MRI-optimized VAE (MR-VAE), which achieves an improvement of approximately 3.92 dB in PSNR for undersampled MRI reconstruction compared to that with SD-VAE \cite{sd}, and (3) Dual-Stage Sampler, a modified version of spaced DDPM sampler, which enforces high-fidelity reconstruction in the latent space. Experiments on the fastMRI dataset\cite{fastmri} demonstrate the state-of-the-art performance of the proposed method and its robustness across various scenarios. The effectiveness of each module is also verified through ablation experiments.