share_log

中国电信:“星辰”大模型在央企率先开源 完成首个全国产化万卡万参大模型训练

CHINA TELECOM: The "Xingchen" large model is the first to be open-sourced among state-owned enterprises, completing the first fully domestically produced model training with ten thousand cards and ten thousand parameters.

Breakings ·  Dec 24 20:10

CHINA TELECOM held the Earnings Conference for the third quarter of 2024, where the company's Executive Director, Chairman, and CEO Ke Ruiwen discussed the AI layout, stating that with the rapid development of information technology, AI has become an important force in a new round of technological revolution and industrial transformation. The company independently developed the "Xingchen" large model capability system covering semantics, speech, vision, and multimodality, being the first to open-source it among state-owned enterprises and completing the first fully domestically produced model training with ten thousand cards and ten thousand parameters, creating the industry's first large model for speech recognition that supports free mixing of 40 dialects. The construction of the "Xinghai" data intelligence platform has formed a high-quality dataset of 9 trillion tokens. The "Lingze" data factor platform has been created to provide full-stack data factor services to over 1,000 partners.

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.
    Write a comment