Ant Group's Zhou Jun: How Hybrid Linear Reform Improves Token Density of Trillion-Parameter Models
Original title:演讲实录 | 从 Token 数量到 Token 密度:万亿参数规模下的高效智能体智能
AI summary
Ant Group Vice President Zhou Jun proposed that a trillion-parameter model's compute cost for every 15 minutes of operation is roughly equivalent to a Tesla, making efficiency the first problem that must be solved in the agent era. The team's hybrid linear reform combining 7 parts Lightning Attention with 1 part MLA reduces the cost of 256K long context from exponential to linear; together with the Kpop algorithm and real-environment training, it cuts token output by about 4 times, enabling a hundred-billion-parameter model to surpass some larger models on real agent tasks.
Key Takeaways
- Ant Group Vice President Zhou Jun proposed that a trillion-parameter model's compute cost for…
- The team's hybrid linear reform combining 7 parts Lightning Attention with 1 part MLA reduces the…
- together with the Kpop algorithm and real-environment training,it cuts token output by about 4…
Quick Facts
- Published
- Updated
- Source
- 公众号:蚂蚁百灵(Ling)
- Category
- Opinion
- AI score
- 60
7:1 的混合注意力配比、将真实错误纳入训练以及区分工具调用与自然语言的奖励函数,为同时追求长上下文效率和智能体可靠性的工作提供了可参照的技术路线。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Ant Group's Zhou Jun: How Hybrid Linear Reform Improves Token Density of Trillion-Parameter Models. July 8, 2026. https://new.nat6.net/en/news/cmsdu2qga010drofz1h40txvh/