Is is possible to auto generate Python APIs?

[aftersnow]

We have several related discussions here: #48 . What do you think? @caozhuozi

[caozhuozi]

Let me investigate!

[caozhuozi]

@aftersnow could you clarify what exactly we mean by the “Python API” in this context? A JSON Schema can indeed be transformed into a Python data structure, but that still feels somewhat distant from what I would typically call a Python API.

What I mean by a Python API here is something that users can actually import and use directly to build and upload AI models.

[aftersnow]

@caozhuozi To build and upload models is the purpose of modctl, but not model-spec. I meant we can provide python data structures for the downstream projects, such as ramalama, which copies the data structure. So python projects can just import model-spec and use it, just like Go.

[github-actions]

This issue is stale because it has been open 40 days with no activity.

[github-actions]

This issue was closed because it has been stalled for 7 days with no activity.

[rishi-jat]

Hi @aftersnow ,

Following the discussion in this issue, I opened a PR to provide Python data structures for the ModelPack specification so downstream Python projects can import spec types directly instead of copying them manually.

PR: #184

The approach in this PR follows the current repository structure:

• The JSON Schema (schema/config-schema.json) is treated as the canonical source, since it is already used for validation.
• Python models are automatically generated from the schema, so they stay synchronized with the specification as it evolves.
• This avoids manually maintaining duplicate structures in another language.
• The implementation keeps the repository minimal and spec-focused, similar to how specs-go provides Go structures for the spec.

This provides Python projects with importable types like: from model_spec.v1 import Model

so they can construct and validate ModelPack configurations using the same schema-defined structure.

Feedback is welcome.