Split the model

why do we need to split the model

Sometimes a model is a little large for a single GPU, we can deploy the model into several GPUS by splitting one node into several nodes.

method

1 use reshard.py here(not work for llama)

2 use this issue for llama70b

posted @ 2023-11-16 03:08  Daze_Lu  阅读(23)  评论(0)    收藏  举报