Skip to content

fix ray engine crashes on multihost #170

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

Merged
merged 1 commit into from
Aug 20, 2024
Merged

fix ray engine crashes on multihost #170

merged 1 commit into from
Aug 20, 2024

Conversation

sixiang-google
Copy link
Collaborator

Fix #150 without all_gather the kv cache

@@ -113,7 +113,7 @@ def prefill_impl(
results = ray.get(all_outputs)
# The prefill function does not return any values;
# the worker itself manages and maintains the prefill states.
return results[0]
Copy link
Collaborator

Choose a reason for hiding this comment

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

Current change will break disaggregation serving. We can still keep this line un-change. Disaggregation do need prefix to be returned in this function.

@@ -490,7 +487,7 @@ def prefill_ray(
samples_per_slot=1,
)

return prefix, result
return result
Copy link
Collaborator

Choose a reason for hiding this comment

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

Let's keep the same function signature between interleave and disaggregated serving. Return None, result instead of return result.

Copy link
Collaborator

@FanhaiLu1 FanhaiLu1 left a comment

Choose a reason for hiding this comment

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

Thanks for making the change, looks good to me.

@sixiang-google sixiang-google merged commit a69ca92 into main Aug 20, 2024
4 checks passed
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.

Ray engine crashes on multihost when fetching Jax.array from prefill_ray
3 participants