FlashQLA: CP-/Bwd-Friendly Fused Linear Attention Kernels for GDN
<style> .katex-display > .katex { font-size: 1.1em; } .katex { font-size: 1.1em; } table .katex { font-size: 1.1em; } </style> Following the release of Qwen3-Next, Gated Delta Network (GDN) has become the workhorse attention layer across the Qwen family — from Qwen3-Next-80B-A...