China Telecom's Xing4.0 MoE goes viral on HF: 29B/4B, trained fully on Ascend 910C
AdinaYakup · x · 2026-09-26
China Telecom's Xing4.0 has been quietly trending on Hugging Face: a 29B/4B MoE trained entirely on Huawei Ascend 910C with MindSpore, 256K context (up to 512K), released under Apache 2.0 — a notable proof point for China's domestic chip training stack.
Related event: China Telecom Open-Sources Xing4.0, Trained Fully on Huawei Ascend Chips(2 posts)→
More from Infra
- SpaceX's Mid-South AI clusters span 2.5M sq ft with millions of GPUs and over 2GW of compute — chaitu · 2026-09-26
- Prime Intellect posts 30 open roles, betting full-stack on open AGI infrastructure — willcb · 2026-09-26
- IIT Delhi researchers unveil India's first indigenous micro-GPU — rvp · 2026-09-26
- Prime Intellect Maps the Emerging AGI Stack: RLaaS, Evals, Inference, Sandboxes and More — willcb · 2026-09-26
- Fireworks x HUD Release RL Training Cookbook: Define Task Once, Train and Evaluate LoRA — sophiamyang · 2026-09-26
- Inside the First Tokenomicon: Amsterdam Event on LLM Token Economics — Bartaseth · 2026-09-26