Skip to content

docs(explainable-palletizer): adapt recipe flow for Cosmos 3#1

Draft
nv-asotelo wants to merge 12 commits into
codex/paula-cookbook-structurefrom
codex/explainable-palletizer-recipe-stage
Draft

docs(explainable-palletizer): adapt recipe flow for Cosmos 3#1
nv-asotelo wants to merge 12 commits into
codex/paula-cookbook-structurefrom
codex/explainable-palletizer-recipe-stage

Conversation

@nv-asotelo

@nv-asotelo nv-asotelo commented Jun 16, 2026

Copy link
Copy Markdown
Owner

Summary

  • Restores the source explainable-palletizer walkthrough shape as the human-facing recipe in cookbooks/cosmos3/explainable-palletizer/README.md.
  • Presents the workflow directly as a Cosmos 3 recipe using a Cosmos3 Reasoner endpoint, a prompt/orchestration client, saved artifacts, and an optional review UI.
  • Keeps runnable prose scoped to generic public components: local images or manifests, /v1/chat/completions, JSON artifacts, and local review from saved files.
  • Keeps the author credit in the README and avoids sensitive runtime details or repeated personal-name references elsewhere.
  • Reworks the supporting markdown files as agent supplements, not competing human recipes.
  • Refreshes assets/main_workflow.svg to show the public input -> client -> Cosmos3 Reasoner -> artifacts/review flow.

Human Entry Point

README.md is the document humans should read first. It contains the complete overview, what runs, why explainable reasoning matters, prerequisites, quickstart, output contract, walkthrough scenarios, results, next steps, resources, and pointers to the appendix files.

The companion files are supplements:

File Role
workflow_e2e.md Scenario and validation appendix for damaged cartons, heavy/rigid goods, mixed-SKU stacking, reasoning modes, and failure signals.
run_custom_images_with_reasoner.md Agent runbook for headless execution with local folders, manifests, mounted files, public dataset materialization, artifacts, and notebook usage.
serve_review_frontend.md Agent runbook for serving a simple browser review UI from saved artifacts.

Section Mapping

Source walkthrough section New location Notes
Title, authors, model/workload table README.md title, credit, and model table Preserves the human opening structure and presents Cosmos3-Nano as the default profile with Cosmos3-Super as the higher-quality profile.
Overview README.md Overview Keeps the palletizing motivation and public Cosmos 3 workflow description.
What You Will Run README.md What You Will Run and Pipeline Components Uses public components: images/manifest, prompt client, Cosmos3 Reasoner endpoint, saved artifacts, optional review UI.
Why Explainable Reasoning README.md Why Explainable Reasoning Keeps the damaged packaging, fragility, load quality, and auditability themes.
Before You Start README.md Before You Start Adds action-first and compact-reasoning modes so agents can choose latency or audit detail.
Prerequisites and Quickstart README.md Prerequisites and Quickstart Documents the endpoint on port 8200, Python client requirements, input data, and output checks.
Model Output Contract README.md Output Contract Keeps bounded actions and fields with Cosmos 3 oriented JSON.
Walkthrough scenarios README.md Walkthrough Scenarios and workflow_e2e.md Keeps the damaged-carton escalation, heavy/rigid lower-layer placement, and mixed-SKU stacking logic.
Results, Next Steps, Resources README.md Results, Next Steps, Resources Keeps the human-facing closeout and points to the agent appendices.

Validation

  • git diff --check
  • Stale-language scan across cookbooks/cosmos3/explainable-palletizer/ for old model family names, old model IDs, adapter names, old service names, old run commands, and stale framing.
  • Contributor-name scan across the recipe path; only the README credit line remains.
  • Local link check among README.md, workflow_e2e.md, run_custom_images_with_reasoner.md, serve_review_frontend.md, and image assets.
  • Extracted and compiled the Python runner snippet from run_custom_images_with_reasoner.md.
  • Extracted and compiled the Python review UI snippet from serve_review_frontend.md.

…rchitecture diagram

Document three control-loop examples (CALL_A_HUMAN, heavy base placement,
mixed-SKU stacking) with UI screenshots, and update the workflow SVG to
show App Server as the central hub.
@nv-asotelo nv-asotelo force-pushed the codex/explainable-palletizer-recipe-stage branch from 23cb499 to 832d1f8 Compare June 16, 2026 14:55
@nv-asotelo nv-asotelo changed the title docs(explainable-palletizer): add scenario walkthroughs and refresh architecture diagram docs(explainable-palletizer): add public Cosmos3 reasoner recipe Jun 23, 2026
`/v1/chat/completions`.
- Local images in `jpg`, `jpeg`, `png`, or `webp` format.

## GPU Requirements

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please use the requirements of the original recipe please. This says nothing to the user.

| --- | --- | --- |
| [Cosmos3-Nano](https://huggingface.co/nvidia/Cosmos3-Nano) Reasoner, [vLLM](https://docs.vllm.ai/) or [Cosmos3 Reasoner NIM](https://catalog.ngc.nvidia.com/orgs/nim/teams/nvidia/containers/cosmos3-reasoner) | Reasoner | Inspect palletizing box images, explain handling risks, and emit bounded action recommendations |

**Original authors:** Kyungchan Son, Minsoo Song, Yujeong Jeong, and Yuri Rocha

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Keep to the original style guides, this is in the body. The credits should be in the header.

@@ -0,0 +1,260 @@
# See How It Thinks: Mixed Palletizing with Explainable Visual Reasoning

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This Readme is an overlay to the original workflow_e2e.md and doesn't help a human reader. ensure everything a user needs - environment setup, credentials, and example use cases along with the graphics are on the front page (README)

@nv-asotelo nv-asotelo changed the title docs(explainable-palletizer): add public Cosmos3 reasoner recipe docs(explainable-palletizer): adapt recipe flow for Cosmos 3 Jun 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant