记住 100 万 token，只存 890 MB

缓存压到每个 token 890 字节，落盘那份只剩上一版的 1/8。做法是大部分层不存自己的缓存，缺了回放 128 个 token 就够。 1 2

#模型架构 #KV 缓存 #DeepSeek #推理优化 #论文解读

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Comments

1/6

2026-10-18

缓存压到每个 token 890 字节，落盘那份只剩上一版的 1/8。做法是大部分层不存自己的缓存，缺了回放 128 个 token 就够。 1 2

#模型架构 #KV 缓存 #DeepSeek #推理优化 #论文解读

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.