FlashRT provides adapters and kernel implementations for robot-policy frameworks, model libraries, and serving runtimes. Use the corresponding integration rather than assuming every host exposes the same API or supports every model.
| Project | Implementation and usage |
|---|---|
| HF Kernels | Kernel packages and source and usage |
| LeRobot, OpenPI, Isaac GR00T | FlashRT-Structures host adapters; standalone Thor reproduction: OpenPI π0.5 and GR00T N1.7 |
| Transformers, Diffusers | FlashRT-Structures adapters and examples, using the FlashRT kernel catalog |
| vLLM, SGLang | FlashRT-Structures serving integrations; recorded demos |
EagleVLA-Edge is PKU-SEC-Lab's llama.cpp-based onboard VLA inference engine. Their C API provider exposes PI0/PI0.5 GGUF inference to FlashRT's Python model interface without starting the foreground HTTP server. See their README for setup and supported interfaces. Their GR00T N1.7 support uses separate HTTP and C APIs; their FlashRT provider currently covers PI0 and PI0.5.