Long context large language models (LLMs) pose significant challenges for efficient serving due to the large memory footprint and high access overhead of KV cache.Retrieval-based KV cache reduction methods can mitigate these challenges, typically by offloading the complete KV cache to CPU and retrieving necessary tokens on demand during inference.However, these methods still suffer from unsatisfactory accuracy degradation and extra retrieval overhead.To address these limitations, this paper proposes A²ATS, a novel retrieval-based KV cache reduction method.A²ATS aims to obtain an accurate approximation of attention scores by applying the vector quantization technique to key states, thereby enabling efficient and precise retrieval of the top-K tokens.First, we propose Windowed Rotary Position Embedding, which decouples the positional dependency from query and key states after position embedding.Then, we propose query-aware vector quantization that optimizes the objective of attention score approximation directly.Finally, we design the heterogeneous inference architecture for KV cache offloading, enabling long context serving with larger batch sizes.Experimental results demonstrate that A²ATS can achieve a lower performance degradation with similar or lower overhead compared to existing methods, thereby increasing long context serving throughput by up to 2.7 ×.

Deploying large language models (LLMs) on edge devices presents significant challenges due to the substantial computational overhead and memory requirements. Activation sparsification can mitigate these resource challenges by reducing the number of activated neurons during inference. Existing methods typically employ thresholding-based sparsification based on the statistics of activation tensors. However, they do not model the impact of activation sparsification on performance, resulting in suboptimal performance degradation. To address the limitations, this paper reformulates the activation sparsification problem to explicitly capture the relationship between activation sparsity and model performance. Then, this paper proposes CHESS, a general activation sparsification approach via CHannel-wise thrEsholding and Selective Sparsification. First, channel-wise thresholding assigns a unique threshold to each activation channel in the feed-forward network (FFN) layers. Then, selective sparsification involves applying thresholding-based activation sparsification to specific layers within the attention modules. Finally, we detail the implementation of sparse kernels to accelerate LLM inference. Experimental results demonstrate that the proposed CHESS achieves lower performance degradation over eight downstream tasks while activating fewer parameters than existing methods, thus speeding up the LLM inference by up to 1.27x.

Large language models (LLMs) exhibit excellent performance in various tasks. However, the memory requirements of LLMs present a great challenge when deploying on memory-limited devices, even for quantized LLMs. This paper introduces a framework to compress LLM after quantization further, achieving about 2.2x compression ratio. A compression-aware quantization is first proposed to enhance model weight compressibility by re-scaling the model parameters before quantization, followed by a pruning method to improve further. Upon this, we notice that decompression can be a bottleneck during practical scenarios. We then give a detailed analysis of the trade-off between memory usage and latency brought by the proposed method. A speed-adaptive method is proposed to overcome it. The experimental results show inference with the compressed model can achieve a 40% reduction in memory size with negligible loss in accuracy and inference speed.