VPP: Virtual Pipeline Parallelism for Efficient Chunked Prefill in Long-Context LLM Inference

2026-08-26

AI Summary

研究提出虚拟管道并行技术(VPP)，通过固定块大小和优化管道布局解决长序列推理中的延迟不平衡问题，提高处理效率。

Abstract

Chunked prefill pipeline parallelism (CPP) is a key technique for LLM inference. However, equal-size chunks exhibit imbalanced latency, as later chunks attend longer prefix KV caches and incur higher attention costs, leading to pipeline bubbles. Existing approaches mitigate this imbalance through dynamic chunk resizing (Dynamic CPP, DCPP), but our measurements show that this trades scheduling overhead for load balancing, which becomes unfavorable on long sequences. In this study, we propose Virtual Pipeline Parallelism (VPP), which keeps chunk sizes fixed and optimizes the pipeline layout through virtual stages. A V-shaped virtual-stage traversal overlaps each chunk's expensive middle stages with the lighter head and tail stages of its neighbors, while asynchronous communication and pipelined packing further reduce communication stalls and cross-request drain bubbles. We implement VPP in vLLM-Ascend and evaluate it on three MoE-based LLMs with sequences up to 1M tokens on 16 Ascend 910C NPUs. VPP improves throughput by up to 13.1% over DCPP on long sequences and 6.7% on mixed workloads, while preserving performance on short sequences. On a 512K-token DeepSeek-V3.1 prefill workload, VPP reduces the pipeline bubble ratio from 6.4% to 0.1%, achieving a 98.0% reduction compared with DCPP.

Problem

• Chunked Prefill
• Pipeline Parallelism
• Latency Imbalance
• Dynamic Chunk Resizing
• Long-Context LLM Inference

Innovation

• Virtual Pipeline Parallelism
• Chunked Prefill
• Pipeline Bubbles
• Asynchronous Communication
• Pipelined Packing