Rotation Q (2 angles), sparse c_proj (2 nonzero), parabolic lm_head, factorized embed, sinusoidal PE (period 11)
DeepSeek 的 15 万次，按任何合理标准来看都是可以忽略的数字。Moonshot 和 MiniMax 合计 1650 万次，量级是另一回事——但能转化成多少真实能力，取决于他们能不能解决「如何用好这些数据」的技术问题。
。关于这个话题，搜狗输入法2026提供了深入分析
numbers, the 4732 is most often referenced but others clearly existed, including
The protests were coordinated on the gaming platform Discord.
FT Videos & Podcasts