Google brings native PyTorch support to vLLM and SGLang on TPU
Google announced native PyTorch backends for vLLM and SGLang inference engines powered by TorchTPU. Built with open-source maintainers, the engines include TPU-specific kernels and KV cache management with TPU Sync. Public benchmarks show state-of-the-art performance and industry-leading cost efficiency on Ironwood TPU for long-context and agentic workloads.
