Replies: 1 comment
-
Hi @frankxyy, There is TP for when we use kernel-injection, it is just that the tensor sharding is done in a slightly different way and in different part of the code. If you are interested, you can follow the logic from here: https://github.com/microsoft/DeepSpeed/blob/master/deepspeed/module_inject/replace_module.py#L354 |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
In engine.py script, I found the code above. It seems that for input with kernel injection switch on, the process of autotp will not be executed. Is there a possibility that the two: autotp and kernel injection can both be utilized?
Beta Was this translation helpful? Give feedback.
All reactions