Junyao Gao, Sibo Liu, Jiaxing Li, Yanan Sun, Hanxin Zhu, Weidong Zhang, Jun Zhang, Cairong Zhao · 2026 · arXiv
DOI: 10.48550/arXiv.2610.30036 · open-access PDF
Meningkatkan panjang konteks perhatian secara signifikan meningkatkan kualitas produksi video streaming, sementara menimbulkan biaya memori dan komputasi yang melarang seiring bertambahnya konteks. metode yang ada mencoba untuk menangkap informasi kontekstual ini melalui pemilihan atau kompresi perhatian, tetapi tetap terbatas oleh representasi terbatas dan tidak memiliki pengawasan tingkat perhatian eksplisit untuk mempertahankan kontribusi historis untuk kueri saat ini. Untuk mengatasi ini, kami memperkenalkan Attn-Forcing, kerangka kerja baru yang mempertahankan kontribusi perhatian penuh konteks untuk produksi video streaming di bawah memori terbatas
Machine translation (Meta M2M100); the English original follows.
Increasing the attention context length significantly improves streaming video generation quality, while incurring prohibitive memory and computational costs as the context grows. Existing methods attempt to capture this contextual information through attention selection or compression, but remain limited by bounded representations and lack explicit attention-level supervision to retain historical contributions to current queries. To address this, we introduce Attn-Forcing, a novel framework that preserves full-context attention contributions for streaming video generation under bounded memory
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.30036, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange