Closed peterdudfield closed 3 months ago
This is odd, because the CI test do not catch this but, but only when run in docker
Tests are run with python3.12 and docker is 3.10, I will try to upgrade and see if it helps
This did not help
Could try rolling back torch upgrades, or perhpas look at hydra versions
Althought hydra seems like the same version in each
It might be that there's a mistake in a different package, should be m=4
torch | lightning | CI test | docker run |
---|---|---|---|
2.4.0 | 2.3.3 | yes | no |
2.3.0 | 2.3.3 | yes | TODO |
2.2.0 | 2.3.1 | no | yes |
2.4.0 | 2.3.1 |
error in CI is due to error in app, is due to loading model
Difference between 2.3.20 and 2.3.11
old Cartopy-0.23.0 PVNet-summation-0.3.0 PyYAML-6.0.1 geopandas-1.0.1 huggingface-hub-0.20.3 nowcasting_datamodel-1.5.45 numpy-1.26.4 ocf_datapipes-3.3.35 pvnet-3.0.47 pyogrio-0.9.0 pytorch-lightning-2.1.3 torch=2.2.0
new artopy-0.23.0 PVNet-summation-0.3.2 beautifulsoup4-4.12.3 bs4-0.0.2 fiona-1.9.6 geopandas-0.14.4 huggingface-hub-0.24.1 nowcasting_datamodel-1.5.52 numpy-2.0.0 ocf_datapipes-3.3.39 pillow-10.4.0 pvlive-api-1.3.0 pvnet-3.0.50 pytorch-lightning-2.3.3 pyyaml-6.0.1 soupsieve-2.5 torchvision-0.19.0 torch=2.4.0
when running docker locally, i cant import import pvnet.models.multimodal.multimodal
as i get a torchvision error
COuld try wrapping this in a try
and except
- https://github.com/openclimatefix/PVNet/blob/main/pvnet/models/multimodal/multimodal_base.py#L3, so that the import works
Ok - this is due to a torch vision import error - https://github.com/openclimatefix/pvnet_app/actions/runs/10157687269
This is solved in 2.3.23
#114
Describe the bug
A clear and concise description of what the bug is.
This happens in pvnet_app = 2.3.21 but not in 2.3.20
Expected behavior
Work with no errors
Additional context
We upgraded torch, and pytorch lightning version in this release