You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently individual models may receive additional arguments using AIExtension.model_parameters. This only works if one wants to customize one specific model, not all models of a given provider (e.g. setting a proxy URL for all of them).
Proposed Solution
Add AIExtension.provider_parameters with precedence lower than AIExtension.model_parameters.
Additional context
None
The text was updated successfully, but these errors were encountered:
@krassowski I feel like this issue would be solved if we simply allowed glob expressions matching model IDs to be specified in the allowlist, denylist, and as keys in model_parameters. That way, we don't need 2 traitlets for every configuration field (one model-scoped and one provider-scoped), which I think could be confusing for operators.
I think it's important to recognize that operator personas (e.g. sysadmins) are often less experienced with Jupyter AI than user personas, so ideally, we shouldn't require them to understand the distinction between providers & models when configuring their deployments.
What do you think about implementing model ID globs for traitlets in v3? I can open an issue for that if it seems like a good feature to have in v3.
Problem
Currently individual models may receive additional arguments using
AIExtension.model_parameters
. This only works if one wants to customize one specific model, not all models of a given provider (e.g. setting a proxy URL for all of them).Proposed Solution
Add
AIExtension.provider_parameters
with precedence lower thanAIExtension.model_parameters
.Additional context
None
The text was updated successfully, but these errors were encountered: