Workflow
科大讯飞加码自主可控大模型训练 拟投入24亿元用于算力租赁项目

Group 1 - Company plans to invest 2.4 billion yuan in computing power leasing projects to enhance its autonomous and controllable computing foundation [1] - The adjustment in fundraising usage includes 800 million yuan for working capital and 2.4 billion yuan for computing power leasing [1] - The computing power leasing project is essential for the continuous growth and competitiveness of the company's research and various business lines [1][2] Group 2 - The company launched the "Feixing No. 1" platform, supporting trillion-parameter model training, and plans to train the first domestic billion-parameter model by January 30, 2024 [2] - The computing power leasing project is expected to meet the increasing demands for computing power as large models evolve, with each token input requiring 6 to 8 floating-point operations per parameter [2] - The company has served 806,000 enterprise users with its large model, covering over 300 application scenarios, and sufficient autonomous computing power will lower the customization threshold for industry models [2] Group 3 - The investment in computing power reflects the company's commitment to autonomous controllable technology and strategic investment in the AI industry [3] - The continuous strengthening of the computing power foundation is expected to lead to breakthroughs in more core areas for the company's large model [3] - The company emphasizes the importance of autonomous innovation and competitive advantages in key areas to seize strategic development opportunities [3]