Tencent Open-Sources SAPS Sparse Attention Method for Qwen3

Tencent released Simple Attention Sparsification (SAPS) on Hugging Face, a sparse attention method that learns KV chunking directly from the language modeling loss, fine-tuned on Qwen3-14B to extend context length.

2026-09-14 ~ 2026-09-15 · 2 related posts

• Tencent releases Simple-Attention-Sparsification: sparse attention for long-context Qwen3-14B — tencent · 2026-09-14
• Tencent releases SAPS: sparse attention that learns KV block selection end-to-end — teortaxesTex · 2026-09-15