DeepSeek V4.1 Gives Prefill and Decode Different Compute Paths
This is a dev post classified by Jev as Other (a tutorial), kept by the Dev Radar because it carries real work, not commentary.
DeepSeek V4.1 Gives Prefill and Decode Different Compute Paths Prompt tokens mostly traverse 20 layers; generated tokens traverse all 40. This Causal Encoder-Decoder design nearly halves long-input prefill while preserving autoregressive generation. Zhihu contributor 潜龙勿用, Changxin Ke(柯昌鑫), a graduate researcher at ICT, CAS, explains how its architecture and post-training were designed together. 1️⃣ A causal encoder, not T5 The 40-layer backbone is split into a 20-layer causal encoder and a 20-layer decoder. Both remain causal. The encoder processes the prompt and supplies the decoder’s globa
Posted by Zhihu Frontier (12.1k followers) 2 days ago · 41 likes · 2.5k views · view the original post on X. Kept by the Dev Radar as Other.
More dev work like this
- We’ve evaluated NVIDIA GR00T N1.7 on VLA-Replica! Check out Alex @alexhuang1029's post… — @YuXiang_IRVL
- You can run locally Uncensored Ternary Bonsai 2-27B Same 5.9GB file. — @0x0SojalSec
- dear rtx 3060 owners, and every 12gb card behind it. bonsai 2 27b dense went from 26 to… — @sudoingX
- Nice: banked codex reset for all of us. — @kimmonismus
- Nvidia dropped an official DeepSeek-V4.1-Flash NVFP4 build. — @TeksEdge
- Convolutional Neural Networks. #BigData #Analytics #DataScience #AI #MachineLearning… — @gp_pulipaka
- Reinforcement Learning and Optimal Control! #BigData #Analytics #DataScience #AI… — @gp_pulipaka
- The #Mathematical #Programming - #FPGA. #BigData #Analytics #DataScience #AI… — @gp_pulipaka
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 15.1k posts from 4.7k X accounts over the last 21 days, 1.7k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-19 22:52 UTC. Full method.