This guide explains to optimum-tpu users the difference between Jetstream Pytorch and PyTorch XLA as those are two available backends in TGI.
JetStream PyTorch is a high-performance inference engine built on top of PyTorch XLA. It is optimized for throughput and memory efficiency when running Large Language Models (LLMs) on TPUs.
| Feature | Jetstream Pytorch | PyTorch XLA |
|---|---|---|
| Training | ❌ | ✅ |
| Serving | ✅ | ✅ |
| Performance | Higher serving performance | Standard performance |
| Flexibility | Limited to serving | Full PyTorch ecosystem |
| Use Case | Production inference | Development and training |
| Integration | Optimized for deployment | Standard PyTorch workflow |
Notes: By default, optimum-tpu is using PyTorch XLA for training and Jetstream Pytorch for serving.
You can configure optimum-tpu to use either version for serving with TGI. You can use the Pytorch XLA backend in TGI by setting up -e JETSTREAM_PT_DISABLE=1 in your docker run arguments.
You can find more information about: