train_gpt_demo.py例子中的一些疑问 #2182
Answered
by
ver217
yhcc
asked this question in
Community | Q&A
-
hello,能麻烦帮我解答下关于在ZERO中使用tensor_parallelize的疑问吗? ColossalAI/examples/language/gpt/train_gpt_demo.py Lines 169 to 174 in 65f56f4 这个例子中,似乎只需要对column进行切分,但这样其实是会导致在每个卡上计算出来的tensor是在最后一个维度为1/n_proc原始大小,这里是CAI会做特定的处理让它参与运算后返回的activation是已经gather过,所以大小也是expected的大小了吗? (2)对应的 ColossalAI/examples/language/gpt/train_gpt_demo.py Lines 157 to 168 in 65f56f4 这个161行意思是就是指因为后续还有一个根据row切分的,所以暂时不需要gather回未切分前tensor的大小么? |
Beta Was this translation helpful? Give feedback.
Answered by
ver217
Dec 23, 2022
Replies: 1 comment 1 reply
-
目前为了兼容性是都会自动gather output的 |
Beta Was this translation helpful? Give feedback.
1 reply
Answer selected by
yhcc
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
目前为了兼容性是都会自动gather output的