FlashQLA: CP-/Bwd-Friendly Fused Linear Attention Kernels for GDN

Alibaba's Qwen team open-sourced FlashQLA, a high-performance linear attention kernel library for GDN (Gated Delta Network) that achieves 2-3x forward and 2x backward speedup over the FLA Triton kernel across multiple scenarios on NVIDIA Hopper.

From the source
Today we officially open-source FlashQLA : a high-performance linear attention kernel library built on TileLang . FlashQLA applies reasonable operator fusion and performance optimization to the forward and backward passes of GDN Chunked Prefill, achieving 2-3× forward speedup and 2× backward speedup over the FLA Triton kernel across multiple scenarios on NVIDIA Hopper.
qwen.ai

Visit source ↗qwen.ai

Announced: 28 Apr 2026 02:00 UTC
Detected: 28 Aug 2026 13:23 UTC (+122d after)

Other stories

xAI · 29 AugGrok Bot now works with XCapability change

GitHub · 28 AugGitHub Copilot in Visual Studio — August update

GitHub · 28 AugGitHub Copilot weekly releases — August 24Capability change

Amazon · 28 AugBatch write and discover records in Amazon SageMaker Feature StoreAPI release

Perplexity · 28 AugGLM 5.3

API release

Amazon · 28 AugHow Decathlon runs demand forecasting at scale with Chronos-2