Python Version Compatibility and Dependency Issues in Official Colab Example

[rtot42]

Hi,

I’m trying to run the official Colab notebook provided in the repository for fine-tuning TimesFM, but I’m facing compatibility issues due to Python version requirements and dependency conflicts.

Key Issues:
Python Version Mismatch:

TimesFM requires Python 3.10 to function properly, but some dependencies used for fine-tuning (like Praxis and Lingvo) seem to only support Python 3.9. This creates a fundamental conflict when attempting to run the notebook, especially in environments like Colab, which defaults to Python 3.10.
Dependency Conflicts in Colab:

Installing timesfm upgrades ipython to 8.x, but Colab requires ipython==7.34.0. This results in runtime crashes or a broken environment.
Additional conflicts with pre-installed packages in Colab, such as traitlets and fiddle, make it difficult to replicate the official setup.
Praxis and PaxML Usage:

While the fine-tuning example depends on Praxis and PaxML, these dependencies add significant complexity and incompatibility when setting up the environment. Would it be possible to provide a simplified alternative for fine-tuning that doesn’t rely on these libraries?
Steps Tried:
Manually switching the runtime to Python 3.9. This isn’t feasible in Colab since it defaults to Python 3.10.
Creating a virtual environment. This helps isolate dependencies but isn’t practical within Colab.
Pinning specific package versions to resolve conflicts (e.g., ipython, traitlets), but this hasn’t worked due to runtime limitations in Colab.

[rajatsen91]

We are looking into this issue.

1. I believe paxml and praxis work in 3.10 we have tried this outside colab. But we will try to replicate it in colab.
2. The ipython dependency can be removed from timesfm. We will do that.

[rajatsen91]

For now:

!pip install "git+https://github.com/google-research/timesfm.git@rajat_dev#egg=timesfm[torch]"

works on colab GPU and CPU with python 3.11. We just needed to remove the ipython dependency.