arXiv subjects

Gezi Li

Publications and source records attributed to Gezi Li.

4 recordsLinked to original sources

arXiv2026

FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving

In large language model (LLM) serving systems, the key-value (KV) cache grows linearly with sequence length and concurrency, so long-context inference easily hits the high-bandwidth memory (HBM) capacity wall and results in low concurrency. Offloading the KV cache to host dynamic random-access memory (DRAM) frees HBM space and raises concurrency, but the additional cache management and host-to-device (H2D) load overhead inflates inference latency, higher concurrency fails to translate into effective throughput. We present FlashOffloading, a host-DRAM-centric inference architecture that fully offloads the KV cache in both the prefill and decode phases. FlashOffloading builds rolling HBM cache for prefill, a hot cache for decode, a reverse sparse index with zero extra miss-lookup overhead, and a compute--load overlapped kernel, optimizing KV offloading latency across the entire pipeline and delivering throughput gains under a time-per-output-token (TPOT) service-level objective (SLO). In experiments deploying GLM-5.2 with multi-token prediction (MTP3) speculative decoding on an Ascend910C prefill--decode (PD)-disaggregated platform, FlashOffloading improves prefill throughput by up to 1.38\(\times\) and decode throughput by up to 1.91\(\times\) under a 30 ms TPOT constraint. FlashOffloading is integrated into an open-source serving framework: https://github.com/vllm-project/vllm-ascend/tree/releases/v0.30.0.

cs.AI↗

arXiv2026

EMMA: Event-Driven Multimodal Memory System with Multi-Agent Retrieval

Long video understanding requires reasoning over hours of content that far exceed the context window of multimodal large language models (MLLMs). Existing approaches compress video into fixed-length clips, yet events are fragmented and unrelated content is merged, yielding noisy retrieval. To overcome this challenge, we propose EMMA, an Event-driven Multimodal Memory system with Multi-Agent retrieval, which organizes memory around semantic events in three stages: (1) Event segmentation: a dual-stream detector adaptively locates event boundaries by fusing visual semantic drift with spoken transitions, rather than fixed windows. (2) Memory construction: built on the detected events, a three-level hierarchy of raw video, clip-level memory, and event-level memory stores content at progressively higher semantic abstraction. (3) Memory retrieval: a multi-agent framework collaboratively queries this hierarchy through query decomposition, tiered search, evidence validation, and answer generation, narrowing evidence from coarse events to precise clips. Extensive experiments on three challenging benchmarks demonstrate substantial improvements, showing the effectiveness of our event-aligned memory system. Code, prompt templates, and reproduction scripts are available at https://github.com/BAI-LAB/EMMA.

cs.CV↗

arXiv2026

FFSlim: An Efficient and Lightweight Format for Multi-modal Data Storage and Retrieval

With the rapid expansion of large-scale media-text corpora, multi-modal datasets increasingly require efficient storage and retrieval. Existing formats such as Files, TDP, and FFRecord work adequately for uni-modal data but expose fundamental limitations in multi-modal settings, including storage redundancy, massive small-file overheads, cache-unfriendly layouts, and heavy index structures. These issues jointly inflate storage and memory usage and make I/O the dominant bottleneck in real training workloads. We present FFSlim, a lightweight format for storing and retrieving multi-modal data. FFSlim improves storage efficiency and loading throughput through three components: a unified file format that removes media duplication and avoids small-file proliferation; an adaptive retrieval mechanism that enables low-overhead pair-level access and accelerates repeated media loading; and a redundancy detection and aggregation module that converts existing datasets into the FFSlim layout. The experimental results demonstrate that FFSlim achieves 2.07x and 8.26x higher data loading and write throughput on average than the strongest baseline, with minimal storage and index overhead. Consequently, these underlying I/O accelerations enable FFSlim to reduce end-to-end training time by 5.36%-14.18% across seven diverse multi-modal models.

cs.PF↗

arXiv2025

MILLION: Mastering Long-Context LLM Inference Via Outlier-Immunized KV Product Quantization

Large language models (LLMs) are increasingly utilized for complex tasks requiring longer context lengths, with some models supporting up to 128K or 1M tokens. This trend, however, presents significant challenges in inference speed and memory management. Quantization emerges as a promising approach to address the widening gap between LLM size and memory capacity. However, traditional quantization schemes often yield suboptimal compression results for KV caches due to two key factors: i) On-the-fly quantization and de-quantization, causing significant performance overhead; ii) Prevalence of outliers in KV values, challenging low-bitwidth uniform quantization. To this end, we propose MILLION, a novel quantization framework achieving low-bitwidth KV cache through product quantization. First, we conduct a thorough analysis of KV cache distribution, revealing the limitations of existing quantization schemes. Second, we introduce a non-uniform quantization algorithm based on product quantization, which efficiently compresses data while preserving accuracy. Third, we develop a high-performance GPU inference framework with efficient attention kernel and pipeline design for MILLION that leverages sparse computation and asynchronous quantization, significantly enhancing inference speed. Comprehensive evaluation results demonstrate that MILLION can achieve 4 bits quantization with trivial perplexity and accuracy loss, and achieve 2.09x end-to-end performance gains at 32K context length. Code is released at https://github.com/ZongwuWang/MILLION.

cs.DC↗