cmake: Add ability to pass in LLAMA_BUILD_NUMBER/COMMIT #14167
Conversation
This is useful in the case where llama.cpp is built not from git, but from a tarball, or a distribution package. We move the definitions up earlier in the file, because they must come before add_subdirectory(common) to be effective there.
Both are computed exactly the same way, and it feels like they shouldn't deviate anyway.
There was a problem hiding this comment.
The LLAMA_BUILD_* variables are then also passed on to GGML_BUILD_*, they are both computed the same and I guess they should always be equal anyway when using the embedded ggml build?
I think this works. I guess we could prefix or suffix the ggml number/commit with llama.cpp if needed to disambiguate, but unless we encounter a specific problem with this, no need to do it.
p.s. Use "Squash and merge" when merging the PR
I also limited the squashed messages to their short descriptions, as I saw in some other merges. If I should keep them full in future, let me know. |
|
The body can be as detailed as we want. We just need the commit title to be formatted correctly because the sync scripts rely on this. |
|
I'm thinking we could expand on this and update the CI to push a custom "make dist" tarball that bundles a header/text file with the version info set. That way consumers of the tarball don't need to figure out the version info that was used to bundle it. |
|
@ckastner this might have broken the version numbers in the CMake package. I'm thinking I should probably add a quick CI job now to make sure those remain intact! Something I forgot to do a couple months ago 🤦♂️ |
|
@ckastner It does not break the package! 😊 But I'm adding the CI just to be sure for the future. |
This is useful in the case where llama.cpp is built not from git, but from a tarball, or a distribution package. We already have this for ggml (#1096 over there).
The LLAMA_BUILD_* variables are then also passed on to GGML_BUILD_*, they are both computed the same and I guess they should always be equal anyway when using the embedded ggml build?