This record is derived from the private source snapshot at e450849. It keeps categories and relationships; it does not publish live inventory.
| Domain | Source examples | Evidence boundary |
|---|---|---|
| Access | connect-to-your-spark, Tailscale, VS Code | Names and addresses are operator inputs |
| Multi-system fabric | connect-two-sparks, connect-three-sparks, multi-sparks-through-switch | Cabling and interface maps stay private |
| Frameworks | CUDA-X, JAX, cuTile, PyTorch, NeMo, LLaMA Factory | Versions require fresh checks |
| Serving | vLLM, TensorRT-LLM, SGLang, llama.cpp, Ollama, NIM | Endpoints and model paths are local inputs |
| Applications | Comfy UI, dashboards, VLM, chatbot, RAG | External services and credentials are not published |
| Optimization | NVFP4, speculative decoding, fine-tuning | Results require workload evidence |
The private source also includes Markdown, Python and shell helpers, notebooks, templates, workflow JSON, and media. The public companion records those categories but does not claim every source asset is original or cleared for redistribution.