Skip to content

WIP: Initial draft of a CyclonDX v2.0 AI/ML-BOM guide - #127

Draft
mrutkows wants to merge 1 commit into
CycloneDX:mainfrom
mrutkows:2.0-MLBOM
Draft

mrutkows wants to merge 1 commit into
CycloneDX:mainfrom
mrutkows:2.0-MLBOM

Conversation

@mrutkows

@mrutkows mrutkows commented Nov 9, 2026

Copy link
Copy Markdown
Contributor

First pass at adopting the latest v2.0 (bundled) schema against the existing content of the AI/ML-BOM guide.

Please note we will likely need to refactor all guides to have a new agreed-upon "2.0" pathing scheme to allow v1.7 content to be preserved. I chose to create a new "2.0" directory for now, but a later refectoring PR can be applied to the whole repo. once some consensus is made. Such a discussion can be had here during the comment/review process.

Signed-off-by: Matt Rutkowski <mrutkows@us.ibm.com>
@mrutkows mrutkows self-assigned this Nov 9, 2026
@mrutkows mrutkows added documentation Improvements or additions to documentation t: ML-BOM CDX 2.0 labels Nov 9, 2026
@mrutkows mrutkows changed the title Initial draft of a CyclonDX v2.0 AI/ML-BOM guide WIP: Initial draft of a CyclonDX v2.0 AI/ML-BOM guide Nov 9, 2026
@JunHwan-Kwon

Copy link
Copy Markdown

Thank you, @mrutkows, for putting the 2.0 draft together.

If useful, I would be happy to help with the quantization section in 0x22. The draft currently describes model-level quantization, while the 2.0 schema (CycloneDX/specification#990) also supports it on named input/output parameters. This can be relevant for on-device models whose weights and external I/O use different numeric representations, such as INT8 weights with UINT8 or floating-point I/O.

I could contribute a small input/output quantization example and a short note on how the proposed taxonomy properties in CycloneDX/cyclonedx-property-taxonomy#175 relate to the native 2.0 quantization fields and parameter-level properties.

I can submit this as a PR against your branch or as review comments, whichever is more helpful.

@JunHwan-Kwon

Copy link
Copy Markdown

@mrutkows Thanks for putting the draft together.

I checked the refId/ref consistency point discussed in the meeting, using specification 5a816a3 and guide draft e0925d8.

Across the inspected non-bundled 2.0 schemas, all 83 concrete refId field definitions resolve to refIdType, and all 26 concrete ref field definitions resolve to reference types. No declaration/reference type mismatches were found among those fields.

The complete Appendix C example contains 57 unique declaration IDs, with no duplicates. All 53 local reference occurrences resolve to declared IDs. Selected target-kind checks passed for 33 references; the remaining 20 received existence checks only.

Results, field inventory and reference ledger

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CDX 2.0 documentation Improvements or additions to documentation t: ML-BOM

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants