DeepSeek plans to use Huawei Ascend chips for pre-training its next-generation V5 model
According to sources at the Shanghai Stock Exchange Daily Science Board, DeepSeek plans to pretrain its next-generation V5 model on Huawei Ascend chips. Previously, Ascend handled only DeepSeek's inference and post-training for V4-Pro; the author believes this relates to DeepSeek's planned September 30 open-source of its complete Ascend compute and communication stack (DeepGEMM, FlashMLA, DeepEP).
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original