verifiedhigh · 0.90updated 2026-10-04

FlashDecoding: Splitting long sequences for parallel decoding

paper

Fields

sources/src5-flashdecoding-paper

URL

https://arxiv.org/abs/2311.01918

Publisher

Stanford

Kind

paper

Published

2023-11-01

Accessed

2026-10-04

notes

FlashDecoding splits long sequences into smaller chunks for parallel decoding during the decode phase. Improves GPU utilization for long-context inference. Key technique used in vLLM, FlashAttention.

Notes

FlashDecoding splits long sequences into smaller chunks for parallel decoding during the decode phase. Improves GPU utilization for long-context inference. Key technique used in vLLM, FlashAttention.

Cross-references

This record references no other records.