Skip to content

Revert PR #20793#25120

Closed
nathanmp wants to merge 1 commit into
ggml-org:masterfrom
nathanmp:revert-pr-20793
Closed

Revert PR #20793#25120
nathanmp wants to merge 1 commit into
ggml-org:masterfrom
nathanmp:revert-pr-20793

Conversation

@nathanmp

@nathanmp nathanmp commented Jun 29, 2026

Copy link
Copy Markdown

Overview

Reverts PR #20793 ("Sched: Reintroduce less synchronizations between token, with fixed pipeline parallelism."), which causes a major performance loss on AMD hardware. Johannes said it should be reverted, so here we are.

Requirements

@nathanmp nathanmp requested a review from a team as a code owner June 29, 2026 04:31
@github-actions github-actions Bot added ggml changes relating to the ggml tensor library for machine learning CUDA Related to the CUDA backend labels Jun 29, 2026
@am17an

am17an commented Jun 29, 2026

Copy link
Copy Markdown
Contributor

Please let the author revert PRs.

@am17an am17an closed this Jun 29, 2026
@nathanmp

nathanmp commented Jun 29, 2026

Copy link
Copy Markdown
Author

I read the contribution guidelines and I didn't see anything that prohibits someone other than the author from reverting PRs, so I'm not sure how I should've known that was expected. My bad?

@am17an

am17an commented Jun 29, 2026

Copy link
Copy Markdown
Contributor

It's just common courtesy. We can't put everything in the contribution guidelines

@nathanmp nathanmp deleted the revert-pr-20793 branch June 30, 2026 03:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CUDA Related to the CUDA backend ggml changes relating to the ggml tensor library for machine learning

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants