Yong Li

Haojun Xia, Zhen Zheng, Yuchao Li et al.

With the fast growth of parameter size, it becomes increasingly challenging to deploy large generative models as they typically require large GPU memory consumption and massive computation. Unstructured model pruning has been a common approach to reduce both GPU memory footprint and the overall computation while retaining good model accuracy. However, the existing solutions do not provide a highly-efficient support for handling unstructured sparsity on modern GPUs, especially on the highly-structured Tensor Core hardware. Therefore, we propose Flash-LLM for enabling low-cost and highly-efficient large generative model inference with the sophisticated support of unstructured sparsity on high-performance but highly restrictive Tensor Cores. Based on our key observation that the main bottleneck of generative model inference is the several skinny matrix multiplications for which Tensor Cores would be significantly under-utilized due to low computational intensity, we propose a general Load-as-Sparse and Compute-as-Dense methodology for unstructured sparse matrix multiplication. The basic insight is to address the significant memory bandwidth bottleneck while tolerating redundant computations that are not critical for end-to-end performance on Tensor Cores. Based on this, we design an effective software framework for Tensor Core based unstructured SpMM, leveraging on-chip resources for efficient sparse data extraction and computation/memory-access overlapping. At SpMM kernel level, Flash-LLM significantly outperforms the state-of-the-art library, i.e., Sputnik and SparTA by an average of 2.9x and 1.5x, respectively. At end-to-end framework level on OPT-30B/66B/175B models, for tokens per GPU-second, Flash-LLM achieves up to 3.8x and 3.6x improvement over DeepSpeed and FasterTransformer, respectively, with significantly lower inference cost.

Yuan Yuan, Chenyang Shao, Jingtao Ding et al.

Spatio-temporal modeling is foundational for smart city applications, yet it is often hindered by data scarcity in many cities and regions. To bridge this gap, we propose a novel generative pre-training framework, GPD, for spatio-temporal few-shot learning with urban knowledge transfer. Unlike conventional approaches that heavily rely on common feature extraction or intricate few-shot learning designs, our solution takes a novel approach by performing generative pre-training on a collection of neural network parameters optimized with data from source cities. We recast spatio-temporal few-shot learning as pre-training a generative diffusion model, which generates tailored neural networks guided by prompts, allowing for adaptability to diverse data distributions and city-specific characteristics. GPD employs a Transformer-based denoising diffusion model, which is model-agnostic to integrate with powerful spatio-temporal neural networks. By addressing challenges arising from data gaps and the complexity of generalizing knowledge across cities, our framework consistently outperforms state-of-the-art baselines on multiple real-world datasets for tasks such as traffic speed prediction and crowd flow prediction. The implementation of our approach is available: https://github.com/tsinghua-fib-lab/GPD.

Yi Ding, Chengxuan Tong, Shuailei Zhang et al.

Integrating prior knowledge of neurophysiology into neural network architecture enhances the performance of emotion decoding. While numerous techniques emphasize learning spatial and short-term temporal patterns, there has been limited emphasis on capturing the vital long-term contextual information associated with emotional cognitive processes. In order to address this discrepancy, we introduce a novel transformer model called emotion transformer (EmT). EmT is designed to excel in both generalized cross-subject EEG emotion classification and regression tasks. In EmT, EEG signals are transformed into a temporal graph format, creating a sequence of EEG feature graphs using a temporal graph construction module (TGC). A novel residual multi-view pyramid GCN module (RMPG) is then proposed to learn dynamic graph representations for each EEG feature graph within the series, and the learned representations of each graph are fused into one token. Furthermore, we design a temporal contextual transformer module (TCT) with two types of token mixers to learn the temporal contextual information. Finally, the task-specific output module (TSO) generates the desired outputs. Experiments on four publicly available datasets show that EmT achieves higher results than the baseline methods for both EEG emotion classification and regression tasks. The code is available at https://github.com/yi-ding-cs/EmT.