Sparse Matrix-Matrix Multiplication (SpMM) is a fundamental primitive in scientific computing and artificial intelligence applications. Modern hardware, notably Tensor Core Units (TCUs), offers immense computational power, creating promising opportunities for SpMM acceleration. However, it is difficult to map sparse matrices with irregular structures onto TCUs due to their requirement for regular operands. Existing fixed-granularity tiling methods frequently face a trade-off between massive zero-padding in sparse regions and poor spatial data locality in dense regions. To bridge this gap, we propose TileSpMM, which breaks the static-granularity bottleneck through a variable-size tiling algorithm that dynamically adapts to local sparsity patterns. Furthermore, TileSpMM is equipped with an adaptive load-balancing strategy and customized granularity-specific kernels to improve hardware utilization and mitigate computation redundancy. Experiments on NVIDIA H100 and RTX 5090 GPUs with a diverse range of benchmark matrices show that TileSpMM delivers overall better performance than existing SpMM methods across the evaluated platforms and datasets. Compared with cuSPARSE, SSpMM, Acc-SpMM and FlashSparse, TileSpMM achieves geometric mean speedups of 4.76 × , 2.74 × , 2.38 × and 1.58 × , respectively.
Hongwei Zeng, Shu-Qin Feng, Hao-Cheng Lian et al.· Proceedings of the Internati...· 0 citations
Recent semantic and generative-retrieval recommenders report substantial improvements over ID-only sequential baselines, but it remains unclear whether these gains arise from language-model reasoning, semantic-ID generation, end-to-end semantic architectures, stronger offline item representations, or complementary semantic and collaborative signals. We investigate this attribution ambiguity through LIME-Rec, a lightweight and auditable recovery test. LIME-Rec combines three independent experts: a SASRec sequential expert, an ItemCF co-occurrence expert, and a semantic expert based on frozen BAAI/bge-base-en-v1.5 item embeddings. Their full-catalog scores are normalized per user and combined through auditable score-level fusion followed by bounded history calibration. The fusion gate and calibration head are fitted on validation data only, require no serving-time language-model inference, and keep each expert contribution separately inspectable. On Amazon Beauty, Toys, and Sports, LIME-Rec achieves R@10 scores of 0.0996, 0.1105, and 0.0593, outperforming the strongest comparison baseline by 7.0%-12.0%. Three-expert fusion without history calibration consistently outperforms calibrated SASRec, showing that calibration alone does not explain the recovery. Randomly permuting item-text embeddings across item IDs reduces R@10 by 13.6%-17.5%, indicating that the gains depend on genuine item-text correspondence rather than additional representation capacity. These results suggest that lightweight recovery from offline item representations and transparent fusion should be ruled out before improvements are attributed to serving-time language modeling, semantic-ID generation, or heavier semantic machinery.
Kong Wang, Zhongke He, Xiang Chen et al.· 0 citations
Activation Steering Adapter (ASA), a training-free, inference-time controller that performs a single-shot mid-layer intervention and targets tool domains via a router-conditioned mixture of steering vectors with a probe-guided signed gate to amplify true intent while suppressing spurious triggers is proposed.
Youjin Wang, Run Zhou, Rong Fu et al.· 4 citations· ⚡2
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
