DeepSeek-V3.2-Exp model is officially released and open-sourced
2025-09-29 10:16:07
The DeepSeek-V3.2-Exp model is officially released and open-sourced. The model introduces a sparse Attention architecture, which can effectively reduce computing resource consumption and improve model inference efficiency. At present, the model has been officially put on the Huawei Cloud Large Model-as-a-Service Platform MaaS. For the DeepSeek-V3.2-Exp model, Huawei Cloud still uses the large EP parallel scheme to deploy this time. Based on the sparse Attention structure superposition, the context parallel strategy of long sequence affinity is realized, and model delay and throughput performance are taken into account.
Disclaimer:
1. The information provided does not constitute investment advice. Investors should make independent decisions and bear all risks themselves.
2. The copyright of this content belongs to the original author. The views expressed herein are solely those of the author and do not represent the stance or position of this website.
Previous article：
DeepSeek-V3.2-Exp模型正式发布并开源Next article：
卡塔尔国民银行采用摩根大通的区块链平台处理美元支付
