About training time&memory

[hanxunyu]

Hi, thanks for your great work. May I ask what the training time and memory usage are when using the 7B parameter LLM?
Looking forward to your reply.

[ch3cook-fdu]

For a 7B LLM, it takes about 1 - 2 days and approximately 40GB GPU mem. to achieve reasonable performance. With the help of flash attention, the training process may be even faster. You can make evaluation intervals longer for faster training.