Search papers, labs, and topics across Lattice.
This paper introduces GPUSparse, a GPU-accelerated system for learned sparse retrieval that addresses the CPU-bound bottleneck of traditional inverted index traversal algorithms. By implementing a GPU-parallel inverted index and a batched scatter-add scoring algorithm, GPUSparse achieves exact scoring comparable to CPU methods while delivering a remarkable 235x speedup in query processing. The system demonstrates high throughput and efficiency, reaching 787 queries per second on a large dataset while maintaining superior recall compared to existing CPU solutions.
GPUSparse achieves a staggering 235x speedup over traditional CPU methods while maintaining exact scoring, revolutionizing real-time retrieval efficiency.
Learned sparse retrieval models such as SPLADE achieve retrieval quality competitive with dense models while preserving the interpretability and exact-match advantages of sparse representations. However, inference-time scoring still relies on CPU-bound inverted index traversal algorithms (WAND, Block-Max WAND), creating a fundamental bottleneck for real-time serving at scale. We present GPUSparse, a system for GPU-accelerated exact learned sparse retrieval that introduces: (1) a GPU-parallel inverted index with block-aligned, warp-coalesced posting lists; (2) a batched scatter-add scoring algorithm that processes hundreds of queries simultaneously; and (3) fused Triton kernels with an analysis of the tradeoff between work-efficiency and hardware utilization. On MS MARCO passage ranking (8.8M passages) with real SPLADE embeddings, GPUSparse matches CPU exact scoring to three decimals (MRR@10=0.383, equal to Pyserini SPLADE at this precision; Recall@1000>=0.999 vs. dense matmul, the residual from floating-point tie-breaking) while providing a 235x speedup over Pyserini CPU at 8.8M documents (1.27ms vs. 298ms per query). Compared to Seismic (the fastest CPU sparse retrieval system), which trades 25% recall for speed (R@1000=0.738 vs. 0.983 exact), GPUSparse achieves exact scoring at 787 QPS throughput (batch 500) on the full 8.8M collection, with 1.3ms per query. Our document-parallel kernel reaches 62.6% of H100 peak HBM bandwidth, revealing a fundamental work-efficiency vs. bandwidth-efficiency tradeoff in GPU sparse retrieval. The reformulation of sparse scoring as scatter-add over an inverted index is shared with SPARe's iterative mode; our contribution is its fused-kernel realization, which we measure to be 23-270x faster than a faithful SPARe iterative reimplementation.
MIT CSAIL