CI: add FreeBSD & simplify CUDA windows #3053
Conversation
|
|
||
| - uses: Jimver/cuda-toolkit@v0.2.10 | ||
| id: cuda-toolkit | ||
| with: | ||
| cuda: ${{ matrix.cuda }} | ||
| # TODO(green-sky): _dev seems to fail, and non dev are not enought | ||
| #sub-packages: '["nvcc", "cudart", "cublas", "cudart_dev", "cublas_dev"]' | ||
| sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]' |
There was a problem hiding this comment.
did you get this to work? also, can you create a manual release for this pr in your repo?
There was a problem hiding this comment.
The build passed successfully.
I created a release:
https://github.com/alonfaraj/llama.cpp/releases
I just fixed the folder name and will check it when it done
There was a problem hiding this comment.
There was a problem hiding this comment.
oh god, the cuda runtime for windows grew another 40mb
| @@ -413,6 +396,22 @@ jobs: | |||
| path: | | |||
| cudart-llama-bin-win-cu${{ matrix.cuda }}-x64.zip | |||
|
|
|||
| freeBSD-latest: | |||
| runs-on: macos-12 | |||
There was a problem hiding this comment.
The documentation specify the latest supported macOS version is 12.
https://github.com/cross-platform-actions/action#runners
But it also support Linux, I guess we can try Ubuntu latest.
Do you think it worth change to linux?
There was a problem hiding this comment.
well, it should be fine, since this is an PUBLIC opensource project. If it where PRIVATE, we would have to pay 5x the compute minutes. 😆
i very recently (1 month ago~) got an upgrade to nvidia 525, which provides 12.0 . before that i was on 515 (iirc) which only provided 11.7 . I'm still on ubuntu 20.04 LTS, which is often used on servers. so yea, maybe move to 12.2, but i suggest to also keep the 11.7 around for some time longer. ... also, I should really rework and update my pr to provide linux binaries. |
Yup, that's what I meant. |
Not all changes are code changes. Also, yea, it is somewhat annoying, but it also is supposed to protect from arbitrary code execution (somewhat). |
Green-Sky
changed the title
|
looks like the long running windows cuda builds are currently choking the ci. hope this pr improves it. (despite adding one extra task 😄 ) |
|
@alonfaraj the freebsd ci runner is getting stuck in/after build sometimes for some reason. |
|
see eg https://github.com/ggerganov/llama.cpp/actions/runs/6226790431/job/16900055948 i suggest we implement a global timeout/limit for up to 1h as a save guard. |
* add freebsd to ci * bump actions/checkout to v3 * bump cuda 12.1.0 -> 12.2.0 * bump Jimver/cuda-toolkit version * unify and simplify "Copy and pack Cuda runtime" * install only necessary cuda sub packages
Lllama.cpp allegedly officially support FreeBSD as stated in the README, I think it recommended to cover it in the CI.
This PR:
"Copy and pack Cuda runtime"step to be identical for all CUDA versionsactions/checkoutusages tov3@ggerganov I'm wondering if we should move forward to the latest CUDA - 12.2?