Skip to content

Conversation

@zhenying-liu
Copy link
Contributor

Apply @jax.fused to optimizer state and parameter update. This can avoid memory copies while offloading the optimizer states to the host memory. The device will read and write the host memory directly with fine grained memory control.

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.

3 participants