Skip to content
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

[server] Disable the elastic scheduler when continuous batching is enabled #1569

Merged
merged 2 commits into from
Jan 31, 2024

Conversation

dsikka
Copy link
Contributor

@dsikka dsikka commented Jan 30, 2024

Summary:

  • For this ticket: https://app.asana.com/0/1205229323407165/1206269382450968/f
  • Previously, the server always initialized a Context object as all pipelines used the MultiModelEngine which requires a Context input. Context always uses the elastic scheduler
  • With this PR, for the case where continuous_batch_sizes are provided, a Context object is not used during pipeline initialization and a single stream scheduler is selected.
  • This prevents us from using the MultiModelEngine for this case but any other pipeline initialized without continuous_batch_sizes will use Context/MultiModelEngine

@dsikka dsikka marked this pull request as ready for review January 30, 2024 17:00
@dsikka dsikka requested a review from bfineran January 30, 2024 17:15
@dsikka dsikka merged commit 46ce747 into main Jan 31, 2024
13 checks passed
@dsikka dsikka deleted the update_server branch January 31, 2024 17:49
dsikka added a commit that referenced this pull request Jan 31, 2024
…abled (#1569)

* update server to disable the context/elastic scheduler when continuous batching is enabled

* clean up when context is created
@dsikka dsikka mentioned this pull request Jan 31, 2024
dsikka added a commit that referenced this pull request Feb 1, 2024
* [server] Disable the elastic scheduler when continuous batching is enabled (#1569)

* update server to disable the context/elastic scheduler when continuous batching is enabled

* clean up when context is created

* [TextGeneration] Fix initialization; don't try v1 init for text gen (#1571)

* only check capacity condition durin prefill; already have check in generation

* dont try v1 if running text gen; just raise error
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