Articles

2 articles
Tag: ascend ✕
Attention Editing: Convert LLM Attention Architectures Without Retraining from Scratch ML
2026-04-08T08:31:30.248Z · Src: 2026-04-08T00:00:00.000Z
llmattentionmla
Huawei's Ascend Challenge: New AI Chip 2.87x Faster Than H20, First Chinese FP4 Inference AI
2026-03-22T11:54:17.000Z
Huawei's new Ascend AI chip delivers 2.87x H20 performance with first domestic FP4 inference support, enabling larger LLM deployment at lower cost and reducing China's dependence on NVIDIA.
huaweiascendnvidia