LLama 3 405B had the most unoptimized kv cache usage by far. Deepseek v4 pro use... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

YetAnotherNick 37 days ago | parent | context | favorite | on: Accelerating GPT-5.6 Sol Ultrafast

LLama 3 405B had the most unoptimized kv cache usage by far. Deepseek v4 pro uses 2.4GB for the same context length[1].

[1]: https://vllm.ai/blog/2026-04-24-deepseek-v4

philipportner 37 days ago [–]

Good point, thanks! I haven't been keeping up with most of the new model internals.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact