fix(hipblas): disable avx flags when accellerated bins are used
Description
Seems the hipblas binary still have AVX and other CPU flags that might be problematic on old HW. As hipblas is reserved in cases of running inference on GPU, CPU optimizations aren't needed and actually just causes sigsev when running binaries.
Notes for Reviewers
See: https://github.com/mudler/LocalAI/issues/3367#issuecomment-2480400573
Signed commits
- [ ] Yes, I signed my commits.
Deploy Preview for localai ready!
| Name | Link |
|---|---|
| Latest commit | 8ddd01c818525e14a5ff433b2f7afcddb9d6a7cc |
| Latest deploy log | https://app.netlify.com/sites/localai/deploys/6738a483b31ac600081e586c |
| Deploy Preview | https://deploy-preview-4167--localai.netlify.app |
| Preview on mobile | Toggle QR Code...Use your smartphone camera to open QR code link. |
To edit notification comments on pull requests, go to your Netlify site configuration.
Thanks so much for this @mudler - looking forward to confirming the fix as soon as possible.
@mudler I've tried again with image localai/localai:master-hipblas, which I think should now contain the fix, following merge of #3367
...unfortunately it's the same result. Using docker image inspect localai/localai:master-hipblas to confirm the image ID:
[
{
"Id": "sha256:7c9ffb9f8d249df6d068308adc9e8c1e5b183845031d6ff976b811ffe9106dee",
"RepoTags": [
"localai/localai:master-hipblas"
],
"RepoDigests": [
"localai/localai@sha256:a5dac979e4cac52729619c98daec793537d16c9ecc430d07c1a04cd6628973b7"
],
"Parent": "",
"Comment": "buildkit.dockerfile.v0",
"Created": "2024-11-17T14:46:58.000365484Z",
...
...exit code 132 (Invalid OpCode).
@mudler I've tried again with image
localai/localai:master-hipblas, which I think should now contain the fix, following merge of #3367...unfortunately it's the same result. Using
docker image inspect localai/localai:master-hipblasto confirm the image ID:[ { "Id": "sha256:7c9ffb9f8d249df6d068308adc9e8c1e5b183845031d6ff976b811ffe9106dee", "RepoTags": [ "localai/localai:master-hipblas" ], "RepoDigests": [ "localai/localai@sha256:a5dac979e4cac52729619c98daec793537d16c9ecc430d07c1a04cd6628973b7" ], "Parent": "", "Comment": "buildkit.dockerfile.v0", "Created": "2024-11-17T14:46:58.000365484Z", ......exit code 132 (Invalid OpCode).
I think it didn't contained the fix yet, but it should now - due to https://github.com/mudler/LocalAI/pull/4201. CI was failing before publishing all the images.
To double check you can compare the versions when running LocalAI: if you hit the /version endpoint or the webUI (at the bottom bar) it will show the commit version.