Huawei Releases openPangu-2.0-Pro: A 505B MoE Trained Entirely on Ascend NPUs
zephyr_z9 · x · 2026-07-31
Huawei's flagship open-weight model, openPangu-2.0-Pro, has landed on Hugging Face. It features 505B total parameters, 18B active parameters per token, a 512K context window, and was trained on roughly 34T tokens.
The headline isn't the benchmarks, but the proof that a frontier-scale MoE model can be trained end-to-end entirely outside the Nvidia stack, relying solely on Huawei Ascend NPUs.
Related event: Huawei Open-Sources 505B Parameter Pangu MoE Model(3 posts)→
More from Infra
- Redditor Asks: Why No Video Models Built for 24-32GB Consumer VRAM? — BreadUndPeeTears · 2026-10-16
- Rapid-MLX 0.16.2 brings NVIDIA Nemotron 3.5 up to 120B to Apple Silicon Macs — kuchaev · 2026-10-16
- Adobe ships NVIDIA RTX Spark optimizations: up to 2x faster creative apps and 5x Frame.io gains — shashib · 2026-10-16
- Meta's commitments jumped to $349B in a quarter as server memory prices tripled ahead of Oct 21 earnings — tengyanAI · 2026-10-16
- Meta releases Muse Spark Open, trained with vLLM-powered RL inference from day one — colinmcnamara · 2026-10-16
- a16z: bot 403 blocks double as Amazon cracks down on agentic commerce — a16z Newsletter · 2026-10-16