X · @teortaxesTex
· X / Twitter
RT Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞): Re my point is: if they can train a 1.6T 48B active on 910Cs using all these tricks (that's the i…
RT Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)Re my point is: if they can train a 1.6T 48B active on 910Cs using all these tricks (that's the impressive part, as well as that they wrote so much software around Huawei's immature stack), they can train a 10T 200B on 950DTs. It will simply be done soon.https://x.com/peterwildef