Large language models (LLMs) now support extremely long context windows, but the quadratic complexity of vanilla attention results in significantly long Time-to-First-Token (TTFT) latency. Existing approaches to address this complexity require additional pretraining or finetuning, and often sacrifice model accuracy. In this paper, we first provide both theoretical and empirical foundations for near-lossless sparse attention. We find dynamically capturing head-specific sparse patterns at runtime with low overhead is crucial. To address this, we propose SampleAttention, an adaptive structured and near-lossless sparse attention. Leveraging observed significant sparse patterns, SampleAttention attends to a fixed percentage of adjacent tokens to capture local window patterns, and employs a two-stage query-guided key-value filtering approach, which adaptively select a minimum set of key-values with low overhead, to capture column stripe patterns. Comprehensive evaluations show that SampleAttention can seamlessly replace vanilla attention in off-the-shelf LLMs with nearly no accuracy loss, and reduces TTFT by up to $2.42\times$ compared with FlashAttention.

Exploiting sparsity in deep neural networks (DNNs) has been a promising area to meet the growing computation need of modern DNNs. However, in practice, sparse DNN acceleration still faces a key challenge. To minimize the overhead of sparse acceleration, hardware designers have proposed structured sparse hardware support recently, which provides limited flexibility and requires extra model fine-tuning. Moreover, any sparse model fine-tuned for certain structured sparse hardware cannot be accelerated by other structured hardware. To bridge the gap between sparse DNN models and hardware, this paper proposes tensor approximation via structured decomposition (TASD), which leverages the distributive property in linear algebra to turn any sparse tensor into a series of structured sparse tensors. Next, we develop a software framework, TASDER, to accelerate DNNs by searching layer-wise, high-quality structured decomposition for both weight and activation tensors so that they can be accelerated by any systems with structured sparse hardware support. Evaluation results show that, by exploiting prior structured sparse hardware baselines, our method can accelerate off-the-shelf dense and sparse DNNs without fine-tuning and improves energy-delay-product by up to 83% and 74% on average.

Due to complex interactions among various deep neural network (DNN) optimization techniques, modern DNNs can have weights and activations that are dense or sparse with diverse sparsity degrees. To offer a good trade-off between accuracy and hardware performance, an ideal DNN accelerator should have high flexibility to efficiently translate DNN sparsity into reductions in energy and/or latency without incurring significant complexity overhead. This paper introduces hierarchical structured sparsity (HSS), with the key insight that we can systematically represent diverse sparsity degrees by having them hierarchically composed from multiple simple sparsity patterns. As a result, HSS simplifies the underlying hardware since it only needs to support simple sparsity patterns; this significantly reduces the sparsity acceleration overhead, which improves efficiency. Motivated by such opportunities, we propose a simultaneously efficient and flexible accelerator, named HighLight, to accelerate DNNs that have diverse sparsity degrees (including dense). Due to the flexibility of HSS, different HSS patterns can be introduced to DNNs to meet different applications' accuracy requirements. Compared to existing works, HighLight achieves a geomean of up to 6.4x better energy-delay product (EDP) across workloads with diverse sparsity degrees, and always sits on the EDP-accuracy Pareto frontier for representative DNNs