Unofficial Gumbii Digital project. This repository contains independently maintained notes and workflow playbooks. It is not an NVIDIA product, official documentation set, support resource, or affiliated repository.
I maintain this repository as an evidence-led index of my setup and workflow notes for DGX Spark work. The private source snapshot contains 922 tracked files, generated playbook pages, examples, scripts, and 48 media files. This public branch records what the collection is for, how it was organized, and what I learned while keeping live environment details out of publication.
- The current private baseline is commit e450849; the history shows repeated playbook regeneration.
- The collection covers local access, two- and three-system connectivity, NCCL, CUDA-X data science, JAX, vLLM, llama.cpp, NIM, fine-tuning, dashboards, and related workflows.
- Individual playbooks carry prerequisites, procedure, troubleshooting, and a stopping point.
- Source examples include netplan/configuration examples, notebooks, shell helpers, workflow JSON, and application source.
- Third-party source assets remain governed by their original notices and are not treated as my original work.
| Observation | Source evidence | Status |
|---|---|---|
| 922 tracked files | git ls-files at e450849 |
Historical inventory |
| 48 media assets | extension inventory at e450849 | Historical inventory |
| Repeated "Regenerate all playbooks" commits | git log |
Historical process |
| Dedicated two/three-system and switched-fabric workflows | source playbook directories | Source-backed scope |
A playbook is a contract, not a command dump. It should state prerequisites, read-only checks, owner decisions, acceptance evidence, and recovery.
The source is organized by workflow: access, framework setup, distributed networking, quantization, inference, and applications each have different prerequisites. I keep vendor-specific material separate from my own validation notes and treat topology, model artifacts, and credentials as operator-owned inputs.
See docs/ARCHITECTURE.md for the one public Mermaid boundary.
The counts and commit above are measured from the frozen private snapshot. This branch does not claim every playbook was executed on current hardware. A link or syntax pass proves only that artifact.
- docs/CASE-STUDY.md - collection timeline and operating lessons.
- docs/PLAYBOOK-INVENTORY.md - source-backed domains and asset boundary.
- docs/PUBLICATION-SAFETY.md - publication rules.
- examples/synthetic-playbook-workflow.json - dry-run fixture.
This is a public project interface, not an operational deployment repository. I exclude live addresses, hostnames, hardware identifiers, accounts, local paths, credentials, raw telemetry, service inventories, and actionable topology. Documentation-only aliases use .example names and reserved ranges. Future material is added only after privacy, licensing, and link review. Current compatibility remains an operator preflight responsibility.
Gumbii Digital's original code, documentation, examples, data, diagrams, and media are available under the MIT License. Third-party components, assets, product names, and trademarks retain their respective terms; see COPYRIGHT.md for scope.