Skip to content

[Disco] Add loader for presharded params.#15676

Closed
csullivan wants to merge 2 commits intoapache:unityfrom
csullivan:presharded_loader
Closed

[Disco] Add loader for presharded params.#15676
csullivan wants to merge 2 commits intoapache:unityfrom
csullivan:presharded_loader

Conversation

@csullivan
Copy link
Copy Markdown
Contributor

@csullivan csullivan commented Sep 5, 2023

Support loading params that are already sharded via a rewrite transform and the use of LiftTransformParams.

@csullivan csullivan marked this pull request as ready for review September 7, 2023 17:50
int worker_id = worker->worker_id;
int num_shards = worker->num_workers;
Device device = worker->default_device;
size_t index = weight_index * num_shards + worker_id;
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this line line imply that a sharded parameter set contains only sharded parameters? I don't think that assumption is true for our use case. (e.g. Sharded weights/bias for matmul, but unsharded vocab embedding)

@Lunderberg
Copy link
Copy Markdown
Contributor

In discussion with @csullivan , this PR is superseded by #15957.

@Lunderberg Lunderberg closed this Oct 20, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants