Possible to filter by nationality of model?

[mathomp4]

I'm a new user of llmfit discovering it a few days ago.

However, I work for NASA and we have some limits to what models we can use. At work, this is sort of enforced for me, but for safety's sake, I often use the same caution at home. Juuuuust in case. For example, Deepseek models are explicitly banned. And other Chinese models have similar restrictions. I tend to err on the side of caution and just avoid all of them to be safe.

Is there a way in llmfit to "exclude" those? Even if not built-in, can I supply a "regex" or the like so that if it matches "deepseek" or "qwen" etc. they are not shown?

[AlexsJones]

Thanks for raising the constraint clearly. llmfit does not currently attach provenance or nationality metadata to model results, and it would be unsafe to infer it from a model name or publisher.

The practical option today is an allowlist in the layer that consumes llmfit output, using the model IDs your organisation has approved. If you want an in-tool filter, please open an issue describing the source of truth you would trust for that allowlist and whether it must work offline. A regex blocklist can be a local workaround, but it will be incomplete and easy to evade as names change.

[AlexsJones]

Thanks for raising the constraint clearly. llmfit does not currently attach provenance or nationality metadata to model results, and it would be unsafe to infer it from a model name or publisher.

The practical option today is an allowlist in the layer that consumes llmfit output, using the model IDs your organisation has approved. If you want an in-tool filter, please open an issue describing the source of truth you would trust for that allowlist and whether it must work offline. A regex blocklist can be a local workaround, but it will be incomplete and easy to evade as names change.

[AlexsJones]

One practical refinement: make an allowlist exact model IDs and revisions where your approval process supports that, rather than publisher-name patterns. That gives a consuming tool something auditable to enforce without asserting that a name proves a model’s provenance.

[AlexsJones]

One practical refinement: make an allowlist exact model IDs and revisions where your approval process supports that, rather than publisher-name patterns. That gives a consuming tool something auditable to enforce without asserting that a name proves a model’s provenance.