Inference on my machine differs from other machines I tested

Hi, I trained the model on AWS (GPU instance) for 60K steps and got the model. I then tested it on several GPU/CPU instance and results are consistent. When I deploy it locally on my Ubuntu desktop (CPU only), the inferences are totally off. I tested on AWS GPU instance (p2.xlarge), AWS CPU instance (c5d.4xlarge) and also on Colab. All three show consistent answers for a given context and questions. Only on my desktop the answers are way off. Any inputs as to why this could be happening would help. Thanks!

Reproduce the same context and questions on the Ubuntu CPU desktop and the AWS GPU, AWS CPU, and Colab environments, then compare their inference configurations and outputs. Done means identifying the cause of the discrepancy or documenting the missing environment detail needed to diagnose it.

Written by the indexing model from the issue text.