Skip to content

Conversation

CISC
Copy link
Collaborator

@CISC CISC commented Sep 28, 2025

Unsure if you meant this as a definitive fix, or if there's some other logic that needs to be added?

@CISC
Copy link
Collaborator Author

CISC commented Sep 28, 2025

Oh, too slow, nvm. :)

@CISC CISC closed this Sep 28, 2025
@CISC CISC deleted the cisc/fix-ggml-set-rows-dst-buffer branch September 28, 2025 19:21
result->op = GGML_OP_SET_ROWS;
result->src[0] = b;
result->src[1] = c;
result->src[2] = a; // make sure a gets a backend buffer
Copy link
Member

Choose a reason for hiding this comment

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

The reason it ends up without a buffer is because you applied ggml_set_rows on a tensor that is not used by any other operation in the graph. And since the ggml_set_rows is also missing the dependency due to this bug, in the end nothing in the graph depends on this tensor and the allocator correctly decides to not allocate it.

@github-actions github-actions bot added the ggml changes relating to the ggml tensor library for machine learning label Sep 28, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
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