X · @teortaxesTex
· X / Twitter
Huawei trains, basically, a DeepSeek V3.5, except for minor mHC/ModAttn modifications. Whale Paradigm reigns more supreme than ever. But more importan…
Huawei trains, basically, a DeepSeek V3.5, except for minor mHC/ModAttn modifications. Whale Paradigm reigns more supreme than ever.But more importantly: second large model on Ascends.There is no CUDA moat anymore.Poe Zhao: Huawei's flagship open-weight model is out. openPangu-2.0-Pro landed on Hugging Face: 505B total