🚀 Feature

Add multiple TPU generations to the TPU CI tests. We observe an increase in instances in which TorchXLA CI tests succeed on one TPU generation and not another. Features like Pallas kernels, quantization, etc. lead to surfacing support inconsistencies across various hardware generations. The right way to address this issue is to test our CI on all latest TPU generations including v5p, v5e, Trillium, and beyond. Currently our TPU CI runs on TPU v4.

cc @vanbasten23