This repository provides the necessary instructions to reproduce a specific workload on Google Cloud TPUs. The focus is on reliably achieving a performance metric (e.g. throughput) that demonstrates the combined hardware and software stack on TPUs.
-
./training: instructions to reproduce the training performance of popular LLMs, diffusion, and other models with PyTorch and JAX. -
./inference: instructions to reproduce inference performance. -
./microbenchmarks: instructions for low-level TPU benchmarks such as matrix multiplication performance and memory bandwidth.
Note: This is not an officially supported Google product. This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.