AI caption option...

[XeonG]

Is it not possible to just have the commit generate button for AI, also have a option to just use whatever the current model loaded is set on.

Ie you can setup lmstudio but you have to specify the actual model you want, it's all very rigid, and I'm sure its possible with the openai protocol to get the current model loaded already no? an option to just use that would be good.. less time wasted loading a different model to caption with when there might already be a decent model loaded already.

[love-linger]

Done.

[love-linger]

The configuration for AI model in Preferences dialog has also been removed. SourceGit will remember the last choice of model in each service.

[XeonG]

cool, in that drop down list will it show which model might already be loaded? ie maybe you want to switch to a faster one but maybe one loaded is fine, would help distinguish which model to go for.

[love-linger]

SourceGit use OpenAIModelClient.GetModels() to fetch all available models.

And this is the core logic of OpenAIModelClient.GetModels() in openai/openai-dotnet:

internal virtual PipelineMessage CreateGetModelsRequest(RequestOptions options)
{
 ClientUriBuilder uri = new ClientUriBuilder();
 uri.Reset(_endpoint);
 uri.AppendPath("/models", false);
 PipelineMessage message = Pipeline.CreateMessage(uri.ToUri(), "GET", PipelineMessageClassifier200);
 PipelineRequest request = message.Request;
 request.Headers.Set("Accept", "application/json");
 message.Apply(options);
 return message;
}

And the OpenAIModel only contains: CreatedAt, Id and OwnedBy properties.