-
Notifications
You must be signed in to change notification settings - Fork 12
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WIP] 2d parallelism chapter - tensor parallelism #39
base: main
Are you sure you want to change the base?
Conversation
Discussing tensor parallelism here: https://discuss.pytorch.org/t/tp-fsdp-sync-module-states-cpu-offload/211386/2 |
sharding_strategy=ShardingStrategy.NO_SHARD, | ||
device_mesh=mesh["dp"], |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It seems like if you pass in sync_module_states=True, and HYBRID_SHARD, that it will broadcast from the right process group.
I'd test this out. But it also means we'd need to load the weights on ranks 0-7?
Adds details on tensor parallelism.