Today for AI
HOT RADAR
computeHEAT 0.0°

AI CURATED HOT STORY · 10/22/2026

Qwen Team Releases FlashQLA for High-Performance Linear Attention

1 reports archived0 independent sourcesupdated 10/22/2026, 17:35:35
Synthesis & Latest Updates

Alibaba's Qwen team released FlashQLA, a high-performance GPU operator library designed for Gated Delta Network (GDN) linear attention. By employing hardware-friendly algebraic rewrites and sequence parallelism, it addresses memory redundancy and SFU overload in long sequences, achieving 2-3x speedups in forward and backward passes on Hopper and Blackwell architectures.

LATEST/Alibaba's Qwen team released FlashQLA, a high-performance GPU operator library designed for Gated Delta Network (GDN) linear attention. By employing hardware-friendly algebraic rewrites and sequence parallelism, it addresses memory redundancy and SFU overload in long sequences, achieving 2-3x speedups in forward and backward passes on Hopper and Blackwell architectures.

Hourly heat curve

167 fully observed hours
Now
0.0
Peak
11.918:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. InfoQ 中文T2·78 pts

    Qwen Team Releases FlashQLA for High-Performance Linear Attention

    • FlashQLA is a specialized GPU operator library for GDN linear attention, aiming to break the O(N²) bottleneck of traditional softmax attention.
    • Key techniques include hardware-friendly algebraic rewrites, sequence dimension blocking, and warp specialization co-optimization.
    • Leveraging sliding window properties reduces memory bandwidth pressure, offering out-of-the-box official backend support on mainstream GPU architectures.