Thank you very much! I will indeed inspect it. And yes, when I install them manually it work just fine 🙂

can I see your requirements.txt

I don't know. I never wrote it anywhere :(

Python 3.7.6 (default, Jan 8 2020, 19:59:22) [GCC 7.3.0]

clearml == 0.17.5
langchain == 0.0.27
pandas == 0.24.2
pydantic == 1.8.2
scikit_learn == 0.22.1
tensorboard == 2.5.0
torch == 1.7.1+cu110
transformers == 4.20.1
~atasets == 1.16.1
~qdm == 4.62.3
~umpy == 1.19.5

What is ~atasets==1.16.1 is it suppose to be huggingFace datasets python package with a typo ?

It is it! I will now try and fix it 🙂

I see now that it is wrong there, but is it somthingf that I did wrong?

yes thanks!

I think these messed up packages are taken from a local installation in your python environment (ClearML detects packages used and logs them on the task) - if these are messed up in your local environment, this might affect the resulting task manifest. I'd start by inspecting your local environment using pip freeze and possible taking a look at the dist-packages and site-packages folders (and clean up anything that has a funny-looking name 🙂 )

can you paste the list of collected packages?

can you install those sucessfully correctly ?

or toml