Skip to content

Conversation

am17an
Copy link
Collaborator

@am17an am17an commented Aug 29, 2025

Fix bugs pointed out by @ORippler and @CISC in #15631

  • Removed support for only having norm + add without the multiply
  • Fixed seg fault in op REPEAT because it was trying to access invalid dst->src[I+1], whereas in repeat we only have 1 src
  • Fixed index for add tensor when fusing ops

@github-actions github-actions bot added Nvidia GPU Issues specific to Nvidia GPUs ggml changes relating to the ggml tensor library for machine learning labels Aug 29, 2025
@CISC
Copy link
Collaborator

CISC commented Aug 29, 2025

@am17an Try just ggml-ci, nothing else. :)

ggml-ci
@slaren
Copy link
Member

slaren commented Aug 29, 2025

For ggml-ci to work, the branch needs to be in this repository.

@CISC
Copy link
Collaborator

CISC commented Aug 29, 2025

For ggml-ci to work, the branch needs to be in this repository.

Ahhh.

@slaren
Copy link
Member

slaren commented Aug 29, 2025

The code that triggers the CI is at https://github.com/ggml-org/ci/blob/master/run.sh. It shouldn't be too hard to modify it to monitor all the branches in pull requests using the github API.

@am17an
Copy link
Collaborator Author

am17an commented Aug 29, 2025

Since we can't test the ggml-ci right away, I will proceed to merge this since it fixes some bugs in master. test-backend-ops works fine for me locally and I'll be available in case anything else is required for this PR

@am17an am17an merged commit 8101786 into ggml-org:master Aug 29, 2025
48 checks passed
qnixsynapse pushed a commit to menloresearch/llama.cpp that referenced this pull request Aug 30, 2025
* CUDA: fix bug in rms_norm fusion

* Fix bug for OP_REPEAT

* Fix index for add
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 Nvidia GPU Issues specific to Nvidia GPUs
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants