Zhibin Wang

Yuan Hu, Lei Chen, Zhibin Wang et al.

Data-driven approaches for medium-range weather forecasting are recently shown extraordinarily promising for ensemble forecasting for their fast inference speed compared to traditional numerical weather prediction (NWP) models, but their forecast accuracy can hardly match the state-of-the-art operational ECMWF Integrated Forecasting System (IFS) model. Previous data-driven attempts achieve ensemble forecast using some simple perturbation methods, like initial condition perturbation and Monte Carlo dropout. However, they mostly suffer unsatisfactory ensemble performance, which is arguably attributed to the sub-optimal ways of applying perturbation. We propose a Swin Transformer-based Variational Recurrent Neural Network (SwinVRNN), which is a stochastic weather forecasting model combining a SwinRNN predictor with a perturbation module. SwinRNN is designed as a Swin Transformer-based recurrent neural network, which predicts future states deterministically. Furthermore, to model the stochasticity in prediction, we design a perturbation module following the Variational Auto-Encoder paradigm to learn multivariate Gaussian distributions of a time-variant stochastic latent variable from data. Ensemble forecasting can be easily achieved by perturbing the model features leveraging noise sampled from the learned distribution. We also compare four categories of perturbation methods for ensemble forecasting, i.e. fixed distribution perturbation, learned distribution perturbation, MC dropout, and multi model ensemble. Comparisons on WeatherBench dataset show the learned distribution perturbation method using our SwinVRNN model achieves superior forecast accuracy and reasonable ensemble spread due to joint optimization of the two targets. More notably, SwinVRNN surpasses operational IFS on surface variables of 2-m temperature and 6-hourly total precipitation at all lead times up to five days.

Yilong Chen, Xiang Bai, Zhibin Wang et al.

Video Large Language Models (Video-LLMs) have demonstrated significant potential in the areas of video captioning, search, and summarization. However, current Video-LLMs still face challenges with long real-world videos. Recent methods have introduced a retrieval mechanism that retrieves query-relevant KV caches for question answering, enhancing the efficiency and accuracy of long real-world videos. However, the compression and retrieval of KV caches are still not fully explored. In this paper, we propose \textbf{StreamKV}, a training-free framework that seamlessly equips Video-LLMs with advanced KV cache retrieval and compression. Compared to previous methods that used uniform partitioning, StreamKV dynamically partitions video streams into semantic segments, which better preserves semantic information. For KV cache retrieval, StreamKV calculates a summary vector for each segment to retain segment-level information essential for retrieval. For KV cache compression, StreamKV introduces a guidance prompt designed to capture the key semantic elements within each segment, ensuring only the most informative KV caches are retained for answering questions. Moreover, StreamKV unifies KV cache retrieval and compression within a single module, performing both in a layer-adaptive manner, thereby further improving the effectiveness of streaming video question answering. Extensive experiments on public StreamingVQA benchmarks demonstrate that StreamKV significantly outperforms existing Online Video-LLMs, achieving superior accuracy while substantially improving both memory efficiency and computational latency. The code has been released at https://github.com/sou1p0wer/StreamKV.

Lei Chen, Fei Du, Yuan Hu et al.

Data-driven medium-range weather forecasting has attracted much attention in recent years. However, the forecasting accuracy at high resolution is unsatisfactory currently. Pursuing high-resolution and high-quality weather forecasting, we develop a data-driven model SwinRDM which integrates an improved version of SwinRNN with a diffusion model. SwinRDM performs predictions at 0.25-degree resolution and achieves superior forecasting accuracy to IFS (Integrated Forecast System), the state-of-the-art operational NWP model, on representative atmospheric variables including 500 hPa geopotential (Z500), 850 hPa temperature (T850), 2-m temperature (T2M), and total precipitation (TP), at lead times of up to 5 days. We propose to leverage a two-step strategy to achieve high-resolution predictions at 0.25-degree considering the trade-off between computation memory and forecasting accuracy. Recurrent predictions for future atmospheric fields are firstly performed at 1.40625-degree resolution, and then a diffusion-based super-resolution model is leveraged to recover the high spatial resolution and finer-scale atmospheric details. SwinRDM pushes forward the performance and potential of data-driven models for a large margin towards operational applications.

Zhibin Wang, Shipeng Li, Yuhang Zhou et al.

User experience is a critical factor Large Language Model (LLM) serving systems must consider, where service level objectives (SLOs) considering the experience of individual requests and system level metrics (SLMs) considering the overall system performance are two key performance measures. However, we observe two notable issues in existing metrics: 1) manually delaying the delivery of some tokens can improve SLOs, and 2) actively abandoning requests that do not meet SLOs can improve SLMs, both of which are counterintuitive. In this paper, we revisit SLOs and SLMs in LLM serving, and propose a new SLO that aligns with user experience. Based on the SLO, we propose a comprehensive metric framework called smooth goodput, which integrates SLOs and SLMs to reflect the nature of user experience in LLM serving. Through this unified framework, we reassess the performance of different LLM serving systems under multiple workloads. Evaluation results show that our metric framework provides a more comprehensive view of token delivery and request processing, and effectively captures the optimal point of user experience and system performance with different serving strategies.