Compatability errors between pip packages

[TescaF]

I'm following the robomimic_get_started.ipynb tutorial in Colab, but get the following dependency clashes when trying to run pip install -e robomimic/:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
datasets 4.0.0 requires huggingface-hub>=0.24.0, but you have huggingface-hub 0.23.4 which is incompatible.
peft 0.18.1 requires huggingface_hub>=0.25.0, but you have huggingface-hub 0.23.4 which is incompatible.
gradio 5.50.0 requires huggingface-hub<2.0,>=0.33.5, but you have huggingface-hub 0.23.4 which is incompatible.

It seems like these packages should have specified versions in setup.py to maintain compatibility with huggingface-hub==0.23.4. Could you please share a working list of package versions?

[amandlek]

Does this error prevent installation? It seems like the installation should still have succeeded.

[TescaF]

Technically it doesn't, but it appears to cause some downstream issues. When I follow the rest of the tutorial and run python robomimic/examples/train_bc_rnn.py --debug, I get:

AttributeError: module 'jax.random' has no attribute 'KeyArray'

This appears to be caused by version incompatibilities between JAX, huggingface-hub, and potentially some other packages. I'm running the vanilla robomimic_get_started.ipynb tutorial on Google Colab, so it is not due to pre-existing installations on my machine.

[TescaF]

@amandlek If you have a working setup running in Google Collab, could you share your package versions via pip show?