Post #399260 57 Sep 17, 2026, 02:41 UTC DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compressionhttps://zartbot.github.io/blog/model_arch/dsv41flash_arch/en.html zartbot.github.io DeepSeek-V4.1 Flash: Pushing the Limits of KV Cache Compression · zartbot A deep dive into the DeepSeek-V4.1 Flash technical report: CED, CSA2, HSI, Single-Pass mHC, Engram and FP4 KV Cache — how the KV cache was compressed to just 890 bytes per token.