FRAMEWIREIndonesiaUpdated Sep 1Live wire
0:00 / 0:00

671B MoE. 14GB of VRAM.

16.8 tokens/s decode. That is DeepSeek-V3 Q4_K_M on a single RTX 4090 — not an 8×H100 rack. The hot path stays on the GPU. The giant expert banks live in system RAM. 🆕 KTransformers (kvcache-ai/ktransformers, Apache-2.0, Tsinghua MADSys, SOSP 2025).

EchoGitSep 1
0:00 / 0:00

The shirt that “hides” the person

Remote monitoring Artificial intelligence cameras ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ 🌴It may look like a Hawaiian shirt to you Eye-catching But he is hidden and silent in front of the machine 🤖 ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ 🐎 Every horse has a setback What is the weak point in this innovation?

م. فايز الفرحان Fayez ALanaziSep 1
0:00 / 0:00

DeepSeek Harness Plugins has opened up its capabilities so completely.

DeepSeek Harness Plugins has opened up its capabilities so completely. One advantage is that anyone can combine the human-computer interaction form they want based on the capabilities it opens up. My ideal Skill/App should include both Agent-friendly parts (skill.md) and human-friendly parts (such as interfaces for humans to confirm and read), and the two should be able to communicate with each other.

jo32Sep 1
0:00 / 0:00

Added DeepSeek Flash Vision judge to

Stas KuleshSep 1