Finalize language model provider API

[lramos15]

Does this issue occur when all extensions are disabled?: Yes/No

• VS Code Version:
• OS Version:

Summarizing some discussions surrounding the language model provider API and efforts to finalize it

The idea is to accomplish the following goals

1. Allow extensions to contribute language models
2. Allow users to manage which models appear in the model picker
3. Allow extensions to have custom flows for auth and additioanl model configuration such as Azure Open AI

To complete finalization we should modify the contribution point from

		"languageModels": [
			{
				"vendor": "gemini"
			},
		]

To something like

		"languageModels": [
			{
				"vendor": "gemini",
                                 "displayName": "Google Gemini"
			},
		]

Additionally, we need a way to have a sort of provider provider for the language model lists. The current suggested shape is to follow CallHierarchyProvider and expose a list of language model items that then get passed back to the provide response function

cc @sbatten

[jrieken]

some related API notes 86585d7

[jrieken]

Given LanguageModelChatMessage supports LanguageModelToolResultPart and given the latter supports LanguageModelPromptTsxPart we would need a capability for language models to express if they can handle prompt-tsx or not. In the latter case vscode, could do some to-string transformation

[insilications]

I hope the final API supports some kind of modelOptions that allows the user to specify custom models with custom parameters (temperature, for example). This is sadly missing from the current BYOK experience in Copilot Chat preview.

[jrieken]

It does via LanguageModelChatRequestOptions#modelOptions. Tho, the concrete options are be defined by the corresponding extension and need to picked up by the consumer.