Discepency in results fitting a bambi model with pymc backend locally, compared to remote repository - with same package versions in both environments

Question

I’m working on a Bayesian model using the bambi modeling interface, Pymc backend and numpyro as the inference library.

Problem When I fit my model locally using the following line of code, the resulting coefficients found differ in the order of 0.1 to 0.3% between my local and remote repository:

_fit_output: az.InferenceData = self.estimator.fit(**self.engine_kwargs)

Checks Done

I’ve ensured that all package versions are identical between the local and remote environments. I’ve ensured that the data parsed to the bambi Model are identical between the local and remote environments. I’ve found that my laptop is the only one to have the issue reported, no other collegues face this discrepency.

One thing I noticed is that the inference_library_version in the arviz.InferenceData object is different between my local machine and the remote repository, despite having the same numpyro version installed in both environments.

Here are the details:

Python version: 3.9.11

pymc version: 5.10.3

bambi version: 0.12.0

numpyro version (local and remote): 0.13.2

arviz version: 0.16.1

inference_library_version in arviz.InferenceData object (local): 0.13.2

inference_library_version in arviz.InferenceData object (remote): 0.13.0

I’ve also tried changing the local numpyro version to 0.13.0 and rerunning the model, but the discrepancy persists.

Does anyone have any ideas why this might be happening and how I can resolve it? Any help would be greatly appreciated.

Thank you!