{"slug": "hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models", "title": "HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models", "summary": "Researchers introduced HLA-WM, a hybrid linear attention architecture for long-horizon video world models that combines softmax attention's full-history KV cache with recurrent linear attention's fixed-size state compression to cut memory use. The method targets persistent scene consistency over extended rollouts, where softmax attention's growing KV cache and linear attention's compressed fixed-size states each pose trade-offs.", "body_md": "Long-horizon video world models require persistent memory to preserve scene consistency over extended rollouts. Softmax attention retains the full generation history through a growing KV cache, whereas recurrent linear attention compresses history into fixed-size states with substantially lower memo", "url": "https://wpnews.pro/news/hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models", "canonical_source": "https://aiflash.com/news/131996/", "published_at": "2026-10-06 13:01:28+00:00", "updated_at": "2026-10-06 13:19:50.015314+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "computer-vision", "ai-research", "neural-networks"], "entities": ["HLA-WM"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models", "markdown": "https://wpnews.pro/news/hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models.md", "text": "https://wpnews.pro/news/hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models.txt", "jsonld": "https://wpnews.pro/news/hla-wm-hybrid-linear-attention-for-long-horizon-video-world-models.jsonld"}}