Change protobuf pin in training requirements #13596
Conversation
baijumeswani
added
the
training
| @@ -4,6 +4,6 @@ h5py | |||
| numpy >= 1.16.6 | |||
There was a problem hiding this comment.
Line 509 in 123e1ea
So, when the pipeline does pip install *.whl, it will try to first install all the dependencies listed in setup.py.
|
Thanks for the review @snnn, @askhade, @jywu-msft |
| @@ -4,6 +4,6 @@ h5py | |||
| numpy >= 1.16.6 | |||
| onnx | |||
| packaging | |||
| protobuf >= 3.12.2, <= 3.20.1 | |||
| protobuf >= 3.20.0, <= 3.21.0 | |||
There was a problem hiding this comment.
That doesn't looks right. <= 3.21.0 means you want 3.21.0 (although it won't exist because they call it 4.21.0) but not 3.21.1
There was a problem hiding this comment.
thanks for sharing this. I'll make it < 4.21.1 in that case.
There was a problem hiding this comment.
Better to be <4, because that clearly states "major version introduces breaking change"
There was a problem hiding this comment.
And no it's not <4.21.1, 4.21.0 won't work either.
What happen is, protobuf had a breaking change in 3.21, then its python variant decided to name 3.21 as 4.21 in pip wheel, and there're still new releases of 3.20.x after 4.21.
protobuf 3.18.3 is incompatible with onnx 1.12.0
Some pipelines rely on other sources (example vcpkg) which are older protobuf packages (3.18.3) and as a result, the pipeline fails.
Forcing re-install of protobuf to a newer version.