[codex] Rework NeMo OSS docs orientation

.github/workflows/fern-docs-ci.yml

@@ -0,0 +1,36 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+name: Fern docs (check)
+
+on:
+  pull_request:
+    paths:
+      - 'fern/**'
+      - '.github/workflows/fern-docs-ci.yml'
+
+permissions:
+  contents: read
+
+jobs:
+  check:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v6
+        with:
+          node-version: '22'
+
+      - name: Install Fern CLI
+        run: npm install -g fern-api@$(jq -r .version fern/fern.config.json)
+
+      - name: Validate Fern configuration
+        working-directory: ./fern
+        env:
+          FERN_TOKEN: ${{ secrets.DOCS_FERN_TOKEN }}
+        run: |
+          fern check
+          fern docs md check

.github/workflows/publish-fern-docs.yml

@@ -0,0 +1,42 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Publishes the NeMo OSS hub Fern site.
+#   git tag docs/v0.1.0 && git push origin docs/v0.1.0
+# Requires org secret: DOCS_FERN_TOKEN
+
+name: Publish Fern Docs
+
+on:
+  push:
+    tags:
+      - 'docs/v*'
+  workflow_dispatch: {}
+
+permissions:
+  contents: read
+
+concurrency:
+  group: fern-publish-nemo-framework-hub
+  cancel-in-progress: true
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v6
+        with:
+          node-version: '22'
+
+      - name: Install Fern CLI
+        run: npm install -g fern-api@$(jq -r .version fern/fern.config.json)
+
+      - name: Publish documentation
+        working-directory: ./fern
+        env:
+          FERN_TOKEN: ${{ secrets.DOCS_FERN_TOKEN }}
+        run: fern generate --docs

GH-TOPICS.MD

@@ -0,0 +1,66 @@
+## GitHub Topics Strategy
+
+### Topic Categories
+
+| Category | Topics | Purpose |
+|----------|--------|---------|
+| **Pipeline Stage** | `stage-data`, `stage-training`, `stage-alignment`, `stage-evaluation`, `stage-deployment`, `stage-safety` | Filter by workflow stage |
+| **Model Type** | `model-llm`, `model-vlm`, `model-speech`, `model-diffusion`, `model-omni` | Filter by model modality |
+| **Training Method** | `method-pretraining`, `method-sft`, `method-lora`, `method-rl`, `method-dpo`, `method-grpo` | Filter by training technique |
+| **Backend** | `backend-megatron`, `backend-pytorch`, `backend-vllm`, `backend-tensorrt` | Filter by infrastructure |
+| **Meta** | `nvidia-nemo` | All repos in the framework |
+
+### README stage mapping
+
+Org README columns map to catalog **`stage`** filters in `fern/components/repos.ts`. Optional GitHub topics can mirror these for org-wide filtering:
+
+| README stage | Catalog `stage` | Suggested GitHub topic |
+|--------------|-----------------|------------------------|
+| Data | `data` | `stage-data` |
+| Pretraining | `pretraining` | `stage-training` |
+| RL | `rl` | `stage-alignment` |
+| Inference | `inference` | `stage-evaluation`, `stage-deployment`, `stage-safety` |
+| E2E | `e2e` | — (recipes, pipelines, orchestration) |
+
+Use **`tags`** in `repos.ts` (and topic facets below) for cross-cutting search — e.g. Skills is E2E in the README but also spans SDG and evaluation.
+
+<!-- ### Topic Assignments
+
+| Repo | Pipeline Stage | Model Type | Training Method | Backend |
+|------|----------------|------------|-----------------|---------|
+| **Automodel** | `stage-training` | `model-llm`, `model-vlm`, `model-omni` | `method-pretraining`, `method-sft`, `method-lora` | `backend-pytorch` |
+| **Megatron-Bridge** | `stage-training` | `model-llm`, `model-vlm` | `method-pretraining`, `method-sft`, `method-lora` | `backend-megatron` |
+| **NeMo (Speech)** | `stage-training` | `model-speech` | `method-pretraining`, `method-sft` | `backend-megatron`, `backend-pytorch` |
+| **RL** | `stage-alignment` | `model-llm`, `model-vlm` | `method-sft`, `method-rl`, `method-dpo`, `method-grpo` | `backend-megatron`, `backend-pytorch`, `backend-vllm` |
+| **Gym** | `stage-alignment` | `model-llm`, `model-vlm` | `method-rl` | — |
+| **DFM** | `stage-training` | `model-diffusion` | `method-pretraining` | `backend-megatron`, `backend-pytorch` |
+| **Curator** | `stage-data` | — | — | — |
+| **DataDesigner** | `stage-data` | — | — | — |
+| **Skills** | `stage-data`, `stage-evaluation` | `model-llm` | — | — |
+| **Evaluator** | `stage-evaluation` | — | — | — |
+| **Export-Deploy** | `stage-deployment` | — | — | `backend-vllm`, `backend-tensorrt` |
+| **Guardrails** | `stage-safety` | `model-llm` | — | — |
+| **Run** | `stage-training` | — | — | — |
+| **Emerging-Optimizers** | `stage-training` | — | — | — |
+| **Nemotron** | — | `model-llm` | — | — |
+
+*All repos also get `nvidia-nemo` for org-wide filtering.* -->
+
+### Example Filter Links
+
+Once topics are applied, we you add convenience links to your project's README and docs:
+
+```markdown
+**Browse by pipeline stage:**
+[Data](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:stage-data) ·
+[Training](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:stage-training) ·
+[Alignment](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:stage-alignment) ·
+[Evaluation](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:stage-evaluation) ·
+[Deployment](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:stage-deployment)
+
+**Browse by model type:**
+[LLM](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:model-llm) ·
+[VLM](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:model-vlm) ·
+[Speech](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:model-speech) ·
+[Diffusion](https://github.com/orgs/NVIDIA-NeMo/repositories?q=topic:model-diffusion)
+```

README.md

@@ -1 +1,8 @@
-# .github
+# NVIDIA-NeMo/.github
+
+GitHub organization profile and staged NeMo OSS hub documentation.
+
+- **Org profile** — `profile/README.md` (shown on [github.com/NVIDIA-NeMo](https://github.com/NVIDIA-NeMo))
+- **Hub docs (Fern)** — `fern/` → planned `docs.nvidia.com/nemo/oss` publication target
+
+See [fern/README.md](fern/README.md) for local preview and publish steps.

fern/README.md

@@ -0,0 +1,139 @@
+# NeMo OSS Hub
+
+Fern source for the **NeMo OSS** documentation hub: the lightweight entry point for open source repositories in [NVIDIA-NeMo](https://github.com/NVIDIA-NeMo).
+
+The hub helps readers answer three questions:
+
+- **What should I use?** Task map, lifecycle stages, Framework vs Platform, and the library catalog.
+- **How should I run it?** Runtime chooser, install overview, container catalog, and release notes.
+- **Where do I go next?** Per-library docs, GitHub repos, community links, glossary, and external learning resources.
+
+Per-library docs own commands, APIs, tutorials, model support, and version-specific behavior. This hub owns orientation, routing, stable concepts, and cross-component release metadata.
+
+## Quick Links
+
+| Need | Source | Published Route |
+| --- | --- | --- |
+| Home | [docs/pages/index.mdx](./docs/pages/index.mdx) | `/` |
+| Task-first routing | [docs/pages/get-started/task-map.mdx](./docs/pages/get-started/task-map.mdx) | `/get-started/task-map` |
+| Runtime choice | [docs/pages/get-started/runtime-chooser.mdx](./docs/pages/get-started/runtime-chooser.mdx) | `/get-started/runtime-chooser` |
+| Library catalog page | [docs/pages/about/libraries.mdx](./docs/pages/about/libraries.mdx) | `/about/libraries` |
+| Library catalog data | [components/repos.ts](./components/repos.ts) | Catalog component source |
+| Container catalog page | [docs/pages/about/release-notes/containers.mdx](./docs/pages/about/release-notes/containers.mdx) | `/about/release-notes/containers` |
+| Container catalog data | [components/containers.ts](./components/containers.ts) | Catalog component source |
+| Concepts | [docs/pages/about/concepts/index.mdx](./docs/pages/about/concepts/index.mdx) | `/about/concepts` |
+| Glossary | [docs/pages/resources/glossary.mdx](./docs/pages/resources/glossary.mdx) | `/resources/glossary` |
+| External learning | [docs/pages/resources/external-learning.mdx](./docs/pages/resources/external-learning.mdx) | `/resources/external-learning` |
+| Taxonomy | [TAXONOMY.md](./TAXONOMY.md) | Maintainer reference |
+| GitHub topics guidance | [GH-TOPICS.MD](./GH-TOPICS.MD) | Maintainer reference |
+| Navigation and redirects | [docs.yml](./docs.yml) | Site config |
+
+Published targets are configured in [docs.yml](./docs.yml):
+
+- Preview: `nemo.docs.buildwithfern.com/oss`
+- Production: `docs.nvidia.com/nemo/oss`
+
+## Site Shape
+
+The hub follows the NVIDIA template-library IA, adapted for an ecosystem catalog:
+
+- **About**: overview, ecosystem, architecture, concepts, libraries, release notes.
+- **Get Started**: task map, quickstart, installation, runtime chooser, and stage guides.
+- **Resources**: glossary, external learning, and community.
+
+Concept pages explain durable relationships. The glossary is lookup-oriented. The task map routes user intent to the owning library. The runtime chooser routes setup decisions to containers, installs, source checkout, or Platform setup.
+
+## Content Rules
+
+Keep content here when it is stable and helps a reader choose:
+
+- Framework vs Platform, lifecycle stages, and repo roles.
+- Task-to-library routing.
+- Runtime path decisions.
+- Catalog metadata from `repos.ts` and `containers.ts`.
+- Framework container release metadata and cross-component known issues.
+- Terminology and curated external learning links.
+
+Send content downstream when it changes quickly:
+
+- Install commands, API usage, tutorials, and examples.
+- Model-specific recipes, benchmarks, and support matrices.
+- Version pins and library-only workarounds.
+
+Rule of thumb: add one orienting paragraph and a link instead of copying a library team's material.
+
+## Common Updates
+
+### Add or Update a Repo
+
+Update [components/repos.ts](./components/repos.ts).
+
+- Set `stage` to the org README lifecycle column: `data`, `pretraining`, `rl`, `inference`, or `e2e`.
+- Set `kind` to `library`, `integration`, `reference`, or `infrastructure`.
+- Add durable `tags` for search facets such as modality, technique, or role.
+- Add `docsUrl` and `containerUrl` only when stable public targets exist.
+
+See [TAXONOMY.md](./TAXONOMY.md) for the canonical vocabulary.
+
+### Add or Update a Container
+
+Update [components/containers.ts](./components/containers.ts).
+
+- Keep `latestTag` current.
+- Keep `FRAMEWORK_RECENT_RELEASES` to a small recent set.
+- Add standalone images to `NEMO_CONTAINERS` when they publish.
+- Add cross-component release notes in [known-issues.mdx](./docs/pages/about/release-notes/known-issues.mdx).
+
+### Add a Concept
+
+Add a page under [docs/pages/about/concepts](./docs/pages/about/concepts).
+
+Concept pages should explain relationships, tradeoffs, or decision models. Put short term definitions in [glossary.mdx](./docs/pages/resources/glossary.mdx).
+
+### Add External Learning
+
+Add durable third-party resources to [external-learning.mdx](./docs/pages/resources/external-learning.mdx).
+
+Prefer sources that reveal how external users frame tasks or confusion. Avoid copying commands or version-specific steps.
+
+## Local Development
+
+Prerequisites:
+
+- Node.js 22+
+- Fern CLI: `npm install -g fern-api`
+
+Run checks and preview:
+
+```bash
+cd fern
+fern login
+fern check
+fern docs dev
+```
+
+Open [http://localhost:3000](http://localhost:3000).
+
+Custom React components must be imported in MDX:
+
+```mdx
+import RepoCatalog from "@/components/RepoCatalog";
+
+<RepoCatalog />
+```
+
+## Publish
+
+Publishing uses the NVIDIA Fern organization token: `DOCS_FERN_TOKEN`.
+
+```bash
+git tag docs/v0.1.0 && git push origin docs/v0.1.0
+```
+
+You can also run the **Publish Fern Docs** workflow from GitHub Actions.
+
+## Theme
+
+This site uses `global-theme: nvidia`. Theme assets are owned by [fern-components](https://github.com/NVIDIA/fern-components); keep logos, global CSS, and footer changes there.
+
+The GitHub org link is configured in `navbar-links` and `footer-links` in [docs.yml](./docs.yml). After publishing, verify that the NVIDIA global theme renders the header link.

fern/TAXONOMY.md

@@ -0,0 +1,70 @@
+# NeMo OSS taxonomy
+
+Canonical vocabulary for the staged Fern hub (`docs.nvidia.com/nemo/oss`), org README, and `components/repos.ts`. When copy disagrees, this file wins.
+
+## Top-level map
+
+```
+NVIDIA NeMo (commercial suite — OSS + microservices + NIM + services)
+└── NeMo OSS (GitHub org NVIDIA-NeMo + this hub)
+    ├── NeMo Framework — model lifecycle (libraries + optional NGC bundle)
+    ├── NeMo Platform — agent integration product (CLI, SDK, Studio)
+    └── 22 catalog repos (libraries, integration, reference, infrastructure)
+```
+
+## Terms
+
+| Term | Meaning |
+| --- | --- |
+| **NVIDIA NeMo** | Full software suite spanning open source libraries, commercial products, NIM, microservices, and services. |
+| **NeMo OSS** | Public open source in [NVIDIA-NeMo](https://github.com/NVIDIA-NeMo), staged hub orientation, library documentation, and NGC containers. Entry point for choosing a stack, stage, library, or container. |
+| **NeMo Framework** | Named **model-lifecycle** stack: composable libraries from data through deployment, each with its own source and docs. |
+| **NeMo Framework container** | NGC image `nvcr.io/nvidia/nemo:<tag>`. Bundles Megatron-Bridge, Evaluator, Export-Deploy, Run, and NeMo Speech. |
+| **NeMo Platform** | [nemo-platform](https://github.com/NVIDIA-NeMo/nemo-platform) — CLI, SDK, and Studio for **agent** evaluate / secure / tune / deploy. Composes libraries into an agent integration experience. |
+| **Library** | A focused repo with its own docs and release cadence (Curator, AutoModel, RL, …). |
+| **NeMo Speech** | The [NeMo](https://github.com/NVIDIA-NeMo/NeMo) repo — use this wording for speech AI in user-facing copy. |
+
+## Repo `kind` (catalog metadata)
+
+| Kind | Role | Examples |
+| --- | --- | --- |
+| `library` | Default product repo for a lifecycle stage | Curator, AutoModel, RL, Evaluator |
+| `integration` | Composes multiple libraries into one product surface | NeMo Platform |
+| `reference` | Recipes, cookbooks, reference pipelines | Skills, Nemotron |
+| `infrastructure` | Shared CI or meta repos | FW-CI-templates |
+
+Every repo still has one **stage** (org README columns). **Kind** clarifies role when stage alone is misleading (for example Platform listed under Inference for discoverability).
+
+## Lifecycle stages
+
+Data · Pretraining · RL · Inference · E2E — same columns as [profile/README.md](../profile/README.md).
+
+## Hub page roles
+
+| Page | Job |
+| --- | --- |
+| **Concepts** | Core mental models and relationships. Use concept pages for explanatory topics, not term lookup. |
+| **Ecosystem** | Positioning and choices (Framework vs Platform, commercial boundary) |
+| **Architecture** | Structure — pipeline, backends, containers, Platform overlay |
+| **Libraries** | Inventory from `repos.ts` |
+| **Task map** | Task-first routing from user intent to library, runtime path, and owning docs |
+| **Runtime chooser** | Setup-path decision guide for containers, pip/source installs, and Platform setup |
+| **Glossary** | Lookup-oriented definitions for terms, acronyms, and product names |
+| **External learning** | Curated third-party blogs, videos, and partner examples with freshness caveats |
+
+## Concepts section
+
+Concepts is a directory, not a glossary. Keep pages focused on stable relationships that help readers reason across repos:
+
+| Concept page | Job |
+| --- | --- |
+| **Framework and Platform** | Distinguish model-lifecycle work from integrated agent workflows |
+| **Lifecycle stages** | Explain Data, Pretraining, RL, Inference, and E2E as workflow stages |
+| **Repository catalog model** | Explain stage, kind, and tags |
+| **Training backends and checkpoints** | Explain AutoModel, Megatron-Bridge, and checkpoint flow at a decision level |
+| **Containers and installs** | Explain Framework container, standalone containers, and library installs |
+| **Documentation surfaces** | Explain what the hub, library docs, repos, release notes, and glossary own |
+
+## Broader suite references
+
+Customizer, NIM, and other commercial NeMo microservices have their own product documentation. Mention them only when needed to explain the broader suite; do not use commercial microservice docs as OSS setup destinations.