We're trying different configurations on the hardware side, currently running on A100s via AWS, but will upgrade this soon to get faster inference/training.
I also appreciate you having a free tier, so many other solutions don't.