[RFC] Enable HSDP + CP #463
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This was referenced Jul 16, 2024
Closed
facebook-github-bot
added
the
CLA Signed
wz337
reviewed
Jul 16, 2024
fegin
commented
Jul 17, 2024
| raise NotImplementedError("CP + TP or CP + PP are not supported yet.") | ||
| cp_mesh = world_mesh["cp"] | ||
| dp_mesh = world_mesh["dp"] | ||
| cp_mesh = dp_mesh.reshape((-1, parallel_dims.cp), ("dp", "cp"))["cp"] |
There was a problem hiding this comment.
We may extend ParallelDims to have helpers to get different submeshs. However, the naming can be tricky as there are multiple meaning of dp_mesh.
wanchaol
reviewed
Jul 17, 2024
| assert dp_mesh.mesh_dim_names == ("dp",), dp_mesh.mesh_dim_names | ||
| dp_mesh = dp_mesh.reshape( | ||
| (parallel_dims.dp_replicate, -1), | ||
| ("dp_replicate", "dp_shard"), |
There was a problem hiding this comment.
For HSDP + CP, suppose it's a (2, 2, 2) mesh where HSDP + CP applied, how could CP be used together with HSDP? is CP somewhat need to be merge into one of the HSDP dimensions? i.e. HSDP mesh would be (2, 2), and after merging CP it would become (2, 4) or (4, 2)?
There was a problem hiding this comment.
It would be (2, 4) for fully_shard and it would be (4,) (the merged dimension of first two dimensions of the world_mesh) for data loader.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
Summary:
This PR demonstrates how to use DeviceMesh reshape to enable HSDP + CP