Nvidia CEO Huang says $30 billion OpenAI investment ‘might be the last’
ran on Indigo2...
。关于这个话题，纸飞机官网提供了深入分析
FT Professional
In order to improve this, we would need to do some heavy lifting of the kind Jeff Dean prescribed. First, we could to change the code to use generators and batch the comparison operations. We could write every n operations to disk, either directly or through memory mapping. Or, we could use system-level optimized code calls - we could rewrite the code in Rust or C, or use a library like SimSIMD explicitly made for similarity comparisons between vectors at scale.
for the better.