That was too much, but they still managed to handle over/close to 100 000 docs/rows inserts per second in a batch mode. MongoDB decisively wins here with 115 batch inserts/s vs 81 batch inserts/s for PostgreSQL; latency (queries duration) difference is even more pronounced - 46.766 ms vs 23.828 ms by mean and 127.184 ms vs 821.431 ms by 99th percentile (the lower the better).
但 15 万次是个什么体量？Lambert 认为，这点数据对 DeepSeek 传闻中的 V4 模型或任何模型整体训练的影响可以忽略不计，「更像是某个小团队在内部做实验，大概率连训练负责人都不知道。」。关于这个话题，新收录的资料提供了深入分析
StackSocial prices subject to change.。新收录的资料是该领域的重要参考
DB = "spider.db"。新收录的资料对此有专业解读