Back to Home
RadarAI··Papers & Tech

消息称 DeepSeek 计划用华为昇腾芯片预训练下一代 V5 模型

中文摘要

据传DeepSeek计划用华为昇腾芯片预训练下一代V5模型,近期开源昇腾相关算力栈或与其有关。

English Summary

DeepSeek reportedly plans to pre-train its next-gen V5 model on Huawei Ascend chips, potentially linked to its recent release of Ascend-optimized software stacks.

Original Excerpt

据科创板日报消息人士,DeepSeek 计划在华为昇腾芯片上预训练下一代 V5 模型。此前昇腾只承担 DeepSeek 的推理和 V4-Pro 后训练;作者认为 DeepSeek 于 9 月 30 日开源其完整昇腾算力与通信栈(DeepGEMM、FlashMLA、DeepEP)正与此相关。