| Issue |
56432
|
| Summary |
ML pretrained models are insufficiently opensource.
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
Hello1024
|
A build of LLVM with the ML optimizations enabled will by default include pretrained machine learning models made by Google:
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Analysis/CMakeLists.txt#L8
That URL will be downloaded and included in the binary.
For a FOSS project, it's important that compiled-in files also meet the letter and spirit of the license, and it appears that unfortunately (and perhaps unintentionally) this file doesn't meet that test.
Specifically, it's a binary file built with an unpublished method, using closed source code operating on a closed source dataset. There is available an open source tool, but this file is not produced with it, and it does not produce files with comparable performance.
Details here:
https://github.com/google/ml-compiler-opt/releases/tag/inlining-Oz-v1.1
Calling on LLVM maintainers (and the code author, @jacob-hegna ) to decide the correct course of action to take next.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
