How to `push_to_hub` after training but preserve training logs?

I am currently using an LSF cluster, where my internet access is limited to the master node. However, the job I am running is executed on a work node that does not have internet connectivity. Consequently, I am unable to call the push_to_hub function immediately after the training process ends. Instead, I need to push the trained model to hub to the trainer when I leave the work node and return to the master node. Regrettably, this approach results in the loss of training logs.