Deepseek v4.1 Flash 1017 hitting as high as 72 tok/s decode on repetitive tokens
This is a dev post classified by AGK Intelligence as Other (a launch), kept by the Dev Radar because it carries real work, not commentary.
Deepseek v4.1 Flash 1017 hitting as high as 72 tok/s decode on repetitive tokens 366 GB total (3.9 bpw w/ Engram table, ~2.3 bpw without). 2.25 bpw experts, ~4 bpw dense. Holding 0.045 KLD and 94% top-1 vs. the original FP8 checkpoint per upstream tests. Powered on 2x AMD Gorgons. One @FrameworkPuter @AIatAMD via USB4 on a specialized modified Strix engine. Original engine by @KireiStudios Gorgon modifications by me. So far MiniMax 3.1 Flash and Deepseek V4.1 Flash 1017 both work really well. Next up: Qwen 3.8 Flash Next
Posted by Carlo (7.5k followers) 1 h ago · 20 likes · 777 views · view the original post on X. Kept by the Dev Radar as Other.
More dev work like this
- Just added my signature on the Denim upgrade for the @base security council. — @toady_hawk
- 强化学习做交易，很多人入门撞上的第一个框架就是FinRL，你大概也不例外。 — @bkdgiffug
- With Antioch, you can simulate any deployment environment for your robot — even the… — @antiochrobotics
- took @rehan_shei's open-source pixel fx into a real 3D scene and made the wizard… — @majidmanzarpour
- 彭博终端贵得让个人用户直摇头，这感受你应该不陌生。 — @bkdgiffug
- I went quiet for a few weeks because what I’m building now is better than every library… — @dwhitedesign
- Celestia’s Voyager handled 21M transactions per second in its first big test — @BSCNews
- Our annual kiroween hackathon is here, and submissions are officially open. 🎃 — @kirodotdev
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.2k posts from 5k X accounts over the last 21 days, 8.8k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-10-30 22:12 UTC. Full method.