Original Euangelion schematic; CC BY 4.0. Source facts are attributed; no third-party artwork or trademarks are reproduced.

Asset: oct28-exec-portable-serving-layers.svg
Created: 2026-10-28
Identity: Software layer diagram: llm-d routes to vLLM/SGLang serving engines. One branch passes through TorchTPU to TPUs; another uses GPU backends. Schematic, not a performance comparison.
Source basis:
https://pytorch.org/blog/introducing-torchtpu/
https://developers.googleblog.com/en/announcing-new-vllm-and-sglang-inference-engines-on-tpu/
https://llm-d.ai/blog/llm-d-v1.0-state-of-the-art-inference-ready-for-day-2

Limitations: The new TPU plugins add a hardware path beneath familiar serving engines; llm-d manages requests above them. This simplified dependency diagram does not imply every model, precision mode or configuration is interchangeable.
License: https://creativecommons.org/licenses/by/4.0/