Saved in:
Bibliographic Details
Main Authors: Font, Martí Llopart, Hernando, Javier, España-Bonet, Cristina
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.00028
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • The standard FlashAttention-3 heuristic exhibits a GPU occupancy bottleneck in low-head-count decoding configurations because it disables sequence splitting based on sequence length alone, underutilizing the Streaming Multiprocessors of Hopper GPUs. Our proposed sequence-aware split policy mitigates this by allowing sequence-level parallelism in low-head-count regimes, improving hardware utilization to deliver roughly a 21 to 24% improvement in decoder kernel efficiency on metadata-enabled inference paths, with no observed regressions.