Searcharxiv⌕ Search

arXiv subjects

Shiqiang Nie

Publications and source records attributed to Shiqiang Nie.

4 recordsLinked to original sources

AscSpMM: Efficient SpMM on Ascend NPUs via Pipelined Vector-Cube Collaboration

Neural processing units (NPUs) provide high compute density and energy efficiency through specialized matrix engines and software-managed on-chip memory. However, their libraries primarily target dense computation. Sparse matrix-dense matrix multiplication (SpMM), a key kernel in graph analytics, sparse neural networks, and scientific computing, requires index-driven operand selection, indirect addressing, scattered memory accesses, and load balancing. These requirements conflict with the fixed shapes and contiguous layouts preferred by NPU matrix engines. GPU Tensor Core SpMM methods cannot be directly transferred to NPUs with decoupled vector and matrix engines. We present AscSpMM, which separates irregular operand preparation from regular matrix computation through a parameterized Vector-CSR representation. The vector engine decodes indices and gathers and reorganizes dense operands, while the Cube engine performs matrix multiply-accumulate operations. A double-slot ring buffer enables dependency-aware asynchronous pipelining, while operand reuse and runtime-adaptive tiling reduce cross-engine stalls. On an Ascend 910B3, AscSpMM achieves geometric-mean speedups of 2.36x over GEMM, 19.92x over an existing SpMM implementation, and 5.82x over cuSPARSE on an A100. It reduces matrix-input storage and peak HBM footprint by 48.24% and 22.98% on average, while remaining competitive with specialized GPU Tensor Core SpMM kernels. The code is available at https://github.com/LfLab-AI/Ascend-NPU-SpMM.

cs.AR↗

Adaptive KV Cache Reuse for Fast Long-Context LLM Serving

In long-context Large Language Model (LLM) inference, the Time-To-First-Token (TTFT) latency incurred by the prefill stage has become the foremost bottleneck limiting interactive performance and deployment cost. KV Cache reuse offers a direct path to reduce redundant prefill, yet traditional prefix caching applies only to strict-prefix scenarios; directly reusing KV Cache in non-prefix settings breaks the cross-chunk global attention relationships and causes significant degradation in generation quality. When reusable KV Cache is offloaded to GPU-external cache pools, I/O overheads across heterogeneous hardware tiers further emerge as a new TTFT bottleneck. Efficient non-prefix KV Cache reuse therefore requires both semantic-consistency recovery and compute-I/O co-optimization. This paper presents CacheTune, a frequency-guided and hardware-aware KV Cache reuse system for long-context LLM serving. CacheTune first identifies, offline, the KV pairs most critical to cross-attention recovery through frequency-domain analysis, and then selectively recomputes only these semantic-critical tokens online while reusing the remaining KVs. To turn this semantic selection into end-to-end latency reduction, CacheTune further combines sparse KV transfer, multi-stream asynchronous overlap, deferred positional-encoding recovery, and hardware-aware adaptive recomputation-ratio tuning to balance computation and data movement across heterogeneous cache pools. Evaluations on mainstream LLMs and long-context tasks show that CacheTune achieves 3.72x-4.86x TTFT speedup and 3.93x-6.21x higher throughput while maintaining generation quality close to full recompute. Even when caches are offloaded to I/O-bound SSD/HDD storage, CacheTune sustains 2.34x-2.36x TTFT speedup through adaptive recomputation.

cs.AR↗

Resource-Efficient Speculative Decoding for Long-Context LLM Serving

Speculative decoding reduces sequential Target model calls by verifying multiple tokens from the Draft model in parallel. Yet KV Cache growth limits long-context serving under constrained GPU memory. Offloading KV to CPU memory relieves this pressure. However, existing offloading schemes restore the full KV history before attention and fail to fully exploit the benefits of KV sharing across queries within a verification round. Existing parallel speculative decoding methods also overlook idle GPU compute capacity while memory-bandwidth-bound Target verification waits for historical KV. We present SpecStream, a speculative decoding system that begins verification without waiting for the full KV history to be restored and exploits compute bubbles during KV transfers for concurrent drafting on the same GPUs. It offloads only Target-committed history, keeping candidate rollback local to the GPUs. Each streamed KV chunk serves all queries in the round, with online softmax preserving full attention. Target-priority scheduling controls Draft concurrent execution under resource limits to constrain interference with Target verification. Experiments show that SpecStream maintains task quality close to SGLang speculative decoding while supporting more concurrent requests under limited GPU memory. Across different datasets, it achieves average throughput speedups of 1.41$\times$ and 1.32$\times$ over the offloading baseline for Qwen3 and InternLM2.5, respectively. Compared with parallel speculative decoding on separate Target and Draft GPUs, SpecStream improves output throughput per GPU by an average of 55.4%.

cs.AR↗

KVmix: Gradient-Based Layer Importance-Aware Mixed-Precision Quantization for KV Cache

The high memory demands of the Key-Value (KV) Cache during the inference of Large Language Models (LLMs) severely restrict their deployment in resource-constrained platforms. Quantization can effectively alleviate the memory pressure caused by KV Cache. However, existing methods either rely on static one-size-fits-all precision allocation or fail to dynamically prioritize critical KV in long-context tasks, forcing memory-accuracy-throughput tradeoffs. In this work, we propose a novel mixed-precision quantization method for KV Cache named KVmix. KVmix leverages gradient-based importance analysis to evaluate how individual Key and Value projection matrices affect the model loss, enabling layer-specific bit-width allocation for mix-precision quantization. It dynamically prioritizes higher precision for important layers while aggressively quantizing less influential ones, achieving a tunable balance between accuracy and efficiency. KVmix also introduces a dynamic long-context optimization strategy that adaptively keeps full-precision KV pairs for recent pivotal tokens and compresses older ones, achieving high-quality sequence generation with low memory usage. Additionally, KVmix provides efficient low-bit quantization and CUDA kernels to optimize computational overhead. On LLMs such as Llama and Mistral, KVmix achieves near-lossless inference performance with extremely low quantization configuration (Key 2.19bit Value 2.38bit), while delivering a remarkable 4.9x memory compression and a 5.3x speedup in inference throughput.

cs.LG↗