Adds o3-mini & o1 models

Adds dynamic model loading for GitHub Models
Adds dynamic model loading for Hugging Face
Updates model ordering in the picker

Author: eamodio

CHANGELOG.md

@@ -10,8 +10,10 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/) and this p
 
 - Adds and expands AI model support for GitLens' AI features
   - Adds DeepSeek V3 and R1 models — closes [#3943](https://github.com/gitkraken/vscode-gitlens/issues/3943)
+  - Adds o3-mini and o1 OpenAI models
   - Adds the latest experimental Gemini 2.0 Flash Thinking model
-  - Adds a `gitlens.ai.modelOptions.temperature` setting to specify the temperature (randomness) for AI models
+  - Adds dynamic model loading for GitHub Models and HuggingFace models
+  - Adds a `gitlens.ai.modelOptions.temperature` setting to specify the temperature (randomness) for AI models that support it
   - Adds a _Switch Model_ button to the AI confirmation prompts
 
 ### Changed

docs/telemetry-events.md

@@ -114,7 +114,7 @@
   'failed.error': string,
   'failed.reason': 'user-declined' | 'user-cancelled' | 'error',
   'input.length': number,
-  'model.id': 'claude-3-5-sonnet-latest' | 'claude-3-5-sonnet-20241022' | 'claude-3-5-sonnet-20240620' | 'claude-3-5-haiku-20241022' | 'claude-3-5-haiku-latest' | 'claude-3-opus-latest' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307' | 'claude-2.1' | 'deepseek-chat' | 'deepseek-reasoner' | 'gemini-2.0-flash-exp' | 'gemini-2.0-flash-thinking-exp-01-21' | 'gemini-exp-1206' | 'gemini-exp-1121' | 'gemini-1.5-pro-latest' | 'gemini-1.5-flash-latest' | 'gemini-1.5-flash-8b' | 'gemini-1.0-pro' | 'gpt-4o' | 'gpt-4o-mini' | 'o1-preview' | 'o1-mini' | 'Phi-3.5-MoE-instruct' | 'Phi-3.5-mini-instruct' | 'AI21-Jamba-1.5-Large' | 'AI21-Jamba-1.5-Mini' | 'meta-llama/Llama-3.2-11B-Vision-Instruct' | 'Qwen/Qwen2.5-72B-Instruct' | 'NousResearch/Hermes-3-Llama-3.1-8B' | 'mistralai/Mistral-Nemo-Instruct-2407' | 'microsoft/Phi-3.5-mini-instruct' | 'o1-preview-2024-09-12' | 'o1-mini-2024-09-12' | 'gpt-4o-2024-08-06' | 'gpt-4o-2024-05-13' | 'chatgpt-4o-latest' | 'gpt-4o-mini-2024-07-18' | 'gpt-4-turbo' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview' | 'gpt-4' | 'gpt-4-0613' | 'gpt-4-32k' | 'gpt-4-32k-0613' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-0125' | 'gpt-3.5-turbo-1106' | 'gpt-3.5-turbo-16k' | `${string}:${string}` | 'grok-beta',
+  'model.id': string,
   'model.provider.id': 'anthropic' | 'deepseek' | 'gemini' | 'github' | 'huggingface' | 'openai' | 'vscode' | 'xai',
   'model.provider.name': string,
   'output.length': number,
@@ -133,7 +133,7 @@
   'failed.error': string,
   'failed.reason': 'user-declined' | 'user-cancelled' | 'error',
   'input.length': number,
-  'model.id': 'claude-3-5-sonnet-latest' | 'claude-3-5-sonnet-20241022' | 'claude-3-5-sonnet-20240620' | 'claude-3-5-haiku-20241022' | 'claude-3-5-haiku-latest' | 'claude-3-opus-latest' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307' | 'claude-2.1' | 'deepseek-chat' | 'deepseek-reasoner' | 'gemini-2.0-flash-exp' | 'gemini-2.0-flash-thinking-exp-01-21' | 'gemini-exp-1206' | 'gemini-exp-1121' | 'gemini-1.5-pro-latest' | 'gemini-1.5-flash-latest' | 'gemini-1.5-flash-8b' | 'gemini-1.0-pro' | 'gpt-4o' | 'gpt-4o-mini' | 'o1-preview' | 'o1-mini' | 'Phi-3.5-MoE-instruct' | 'Phi-3.5-mini-instruct' | 'AI21-Jamba-1.5-Large' | 'AI21-Jamba-1.5-Mini' | 'meta-llama/Llama-3.2-11B-Vision-Instruct' | 'Qwen/Qwen2.5-72B-Instruct' | 'NousResearch/Hermes-3-Llama-3.1-8B' | 'mistralai/Mistral-Nemo-Instruct-2407' | 'microsoft/Phi-3.5-mini-instruct' | 'o1-preview-2024-09-12' | 'o1-mini-2024-09-12' | 'gpt-4o-2024-08-06' | 'gpt-4o-2024-05-13' | 'chatgpt-4o-latest' | 'gpt-4o-mini-2024-07-18' | 'gpt-4-turbo' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview' | 'gpt-4' | 'gpt-4-0613' | 'gpt-4-32k' | 'gpt-4-32k-0613' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-0125' | 'gpt-3.5-turbo-1106' | 'gpt-3.5-turbo-16k' | `${string}:${string}` | 'grok-beta',
+  'model.id': string,
   'model.provider.id': 'anthropic' | 'deepseek' | 'gemini' | 'github' | 'huggingface' | 'openai' | 'vscode' | 'xai',
   'model.provider.name': string,
   'output.length': number,
@@ -151,7 +151,7 @@ or
   'failed.error': string,
   'failed.reason': 'user-declined' | 'user-cancelled' | 'error',
   'input.length': number,
-  'model.id': 'claude-3-5-sonnet-latest' | 'claude-3-5-sonnet-20241022' | 'claude-3-5-sonnet-20240620' | 'claude-3-5-haiku-20241022' | 'claude-3-5-haiku-latest' | 'claude-3-opus-latest' | 'claude-3-opus-20240229' | 'claude-3-sonnet-20240229' | 'claude-3-haiku-20240307' | 'claude-2.1' | 'deepseek-chat' | 'deepseek-reasoner' | 'gemini-2.0-flash-exp' | 'gemini-2.0-flash-thinking-exp-01-21' | 'gemini-exp-1206' | 'gemini-exp-1121' | 'gemini-1.5-pro-latest' | 'gemini-1.5-flash-latest' | 'gemini-1.5-flash-8b' | 'gemini-1.0-pro' | 'gpt-4o' | 'gpt-4o-mini' | 'o1-preview' | 'o1-mini' | 'Phi-3.5-MoE-instruct' | 'Phi-3.5-mini-instruct' | 'AI21-Jamba-1.5-Large' | 'AI21-Jamba-1.5-Mini' | 'meta-llama/Llama-3.2-11B-Vision-Instruct' | 'Qwen/Qwen2.5-72B-Instruct' | 'NousResearch/Hermes-3-Llama-3.1-8B' | 'mistralai/Mistral-Nemo-Instruct-2407' | 'microsoft/Phi-3.5-mini-instruct' | 'o1-preview-2024-09-12' | 'o1-mini-2024-09-12' | 'gpt-4o-2024-08-06' | 'gpt-4o-2024-05-13' | 'chatgpt-4o-latest' | 'gpt-4o-mini-2024-07-18' | 'gpt-4-turbo' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview' | 'gpt-4' | 'gpt-4-0613' | 'gpt-4-32k' | 'gpt-4-32k-0613' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-0125' | 'gpt-3.5-turbo-1106' | 'gpt-3.5-turbo-16k' | `${string}:${string}` | 'grok-beta',
+  'model.id': string,
   'model.provider.id': 'anthropic' | 'deepseek' | 'gemini' | 'github' | 'huggingface' | 'openai' | 'vscode' | 'xai',
   'model.provider.name': string,
   'output.length': number,

src/ai/aiProviderService.ts

@@ -1,6 +1,6 @@
 import type { CancellationToken, Disposable, MessageItem, ProgressOptions, QuickInputButton } from 'vscode';
 import { env, ThemeIcon, Uri, window } from 'vscode';
-import type { AIModels, AIProviders, SupportedAIModels, VSCodeAIModels } from '../constants.ai';
+import type { AIProviders, SupportedAIModels, VSCodeAIModels } from '../constants.ai';
 import type { AIGenerateDraftEventData, Sources, TelemetryEvents } from '../constants.telemetry';
 import type { Container } from '../container';
 import { CancellationError } from '../errors';
@@ -33,10 +33,7 @@ export interface AIResult {
 	body: string;
 }
 
-export interface AIModel<
-	Provider extends AIProviders = AIProviders,
-	Model extends AIModels<Provider> = AIModels<Provider>,
-> {
+export interface AIModel<Provider extends AIProviders = AIProviders, Model extends string = string> {
 	readonly id: Model;
 	readonly name: string;
 	readonly maxTokens: { input: number; output: number };
@@ -55,38 +52,39 @@ interface AIProviderConstructor<Provider extends AIProviders = AIProviders> {
 	new (container: Container): AIProvider<Provider>;
 }
 
+// Order matters for sorting the picker
 const _supportedProviderTypes = new Map<AIProviders, AIProviderConstructor>([
 	...(supportedInVSCodeVersion('language-models') ? [['vscode', VSCodeAIProvider]] : ([] as any)),
 	['openai', OpenAIProvider],
 	['anthropic', AnthropicProvider],
-	['deepseek', DeepSeekProvider],
 	['gemini', GeminiProvider],
+	['deepseek', DeepSeekProvider],
+	['xai', xAIProvider],
 	['github', GitHubModelsProvider],
 	['huggingface', HuggingFaceProvider],
-	['xai', xAIProvider],
 ]);
 
 export interface AIProvider<Provider extends AIProviders = AIProviders> extends Disposable {
 	readonly id: Provider;
 	readonly name: string;
 
-	getModels(): Promise<readonly AIModel<Provider, AIModels<Provider>>[]>;
+	getModels(): Promise<readonly AIModel<Provider>[]>;
 
 	explainChanges(
-		model: AIModel<Provider, AIModels<Provider>>,
+		model: AIModel<Provider>,
 		message: string,
 		diff: string,
 		reporting: TelemetryEvents['ai/explain'],
 		options?: { cancellation?: CancellationToken },
 	): Promise<string | undefined>;
 	generateCommitMessage(
-		model: AIModel<Provider, AIModels<Provider>>,
+		model: AIModel<Provider>,
 		diff: string,
 		reporting: TelemetryEvents['ai/generate'],
 		options?: { cancellation?: CancellationToken; context?: string },
 	): Promise<string | undefined>;
 	generateDraftMessage(
-		model: AIModel<Provider, AIModels<Provider>>,
+		model: AIModel<Provider>,
 		diff: string,
 		reporting: TelemetryEvents['ai/generate'],
 		options?: { cancellation?: CancellationToken; context?: string; codeSuggestion?: boolean },
@@ -107,10 +105,10 @@ export class AIProviderService implements Disposable {
 		return this._provider?.id;
 	}
 
-	private getConfiguredModel(): { provider: AIProviders; model: AIModels } | undefined {
+	private getConfiguredModel(): { provider: AIProviders; model: string } | undefined {
 		const qualifiedModelId = configuration.get('ai.model') ?? undefined;
 		if (qualifiedModelId != null) {
-			let [providerId, modelId] = qualifiedModelId.split(':') as [AIProviders, AIModels];
+			let [providerId, modelId] = qualifiedModelId.split(':') as [AIProviders, string];
 			if (providerId != null && this.supports(providerId)) {
 				if (modelId != null) {
 					return { provider: providerId, model: modelId };
@@ -150,10 +148,10 @@ export class AIProviderService implements Disposable {
 	}
 
 	private getOrUpdateModel(model: AIModel): Promise<AIModel | undefined>;
-	private getOrUpdateModel<T extends AIProviders>(providerId: T, modelId: AIModels<T>): Promise<AIModel | undefined>;
+	private getOrUpdateModel<T extends AIProviders>(providerId: T, modelId: string): Promise<AIModel | undefined>;
 	private async getOrUpdateModel(
 		modelOrProviderId: AIModel | AIProviders,
-		modelId?: AIModels,
+		modelId?: string,
 	): Promise<AIModel | undefined> {
 		let providerId: AIProviders;
 		let model: AIModel | undefined;
@@ -552,7 +550,7 @@ export class AIProviderService implements Disposable {
 
 async function confirmAIProviderToS<Provider extends AIProviders>(
 	service: AIProviderService,
-	model: AIModel<Provider, AIModels<Provider>>,
+	model: AIModel<Provider>,
 	storage: Storage,
 ): Promise<boolean> {
 	const confirmed =
@@ -596,9 +594,9 @@ async function confirmAIProviderToS<Provider extends AIProviders>(
 	return false;
 }
 
-export function getMaxCharacters(model: AIModel, outputLength: number): number {
+export function getMaxCharacters(model: AIModel, outputLength: number, overrideInputTokens?: number): number {
 	const tokensPerCharacter = 3.1;
-	const max = model.maxTokens.input * tokensPerCharacter - outputLength / tokensPerCharacter;
+	const max = (overrideInputTokens ?? model.maxTokens.input) * tokensPerCharacter - outputLength / tokensPerCharacter;
 	return Math.floor(max - max * 0.1);
 }
 

src/ai/githubModelsProvider.ts

@@ -1,59 +1,12 @@
+import type { Response } from '@env/fetch';
+import { fetch } from '@env/fetch';
 import type { AIModel } from './aiProviderService';
+import { getMaxCharacters } from './aiProviderService';
 import { OpenAICompatibleProvider } from './openAICompatibleProvider';
 
 const provider = { id: 'github', name: 'GitHub Models' } as const;
 
 type GitHubModelsModel = AIModel<typeof provider.id>;
-const models: GitHubModelsModel[] = [
-	{
-		id: 'o1-preview',
-		name: 'o1 preview',
-		maxTokens: { input: 128000, output: 32768 },
-		provider: provider,
-	},
-	{
-		id: 'o1-mini',
-		name: 'o1 mini',
-		maxTokens: { input: 128000, output: 65536 },
-		provider: provider,
-	},
-	{
-		id: 'gpt-4o',
-		name: 'GPT-4o',
-		maxTokens: { input: 128000, output: 16384 },
-		provider: provider,
-	},
-	{
-		id: 'gpt-4o-mini',
-		name: 'GPT-4o mini',
-		maxTokens: { input: 128000, output: 16384 },
-		provider: provider,
-	},
-	{
-		id: 'Phi-3.5-MoE-instruct',
-		name: 'Phi 3.5 MoE',
-		maxTokens: { input: 134144, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'Phi-3.5-mini-instruct',
-		name: 'Phi 3.5 mini',
-		maxTokens: { input: 134144, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'AI21-Jamba-1.5-Large',
-		name: 'Jamba 1.5 Large',
-		maxTokens: { input: 268288, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'AI21-Jamba-1.5-Mini',
-		name: 'Jamba 1.5 Mini',
-		maxTokens: { input: 268288, output: 4096 },
-		provider: provider,
-	},
-];
 
 export class GitHubModelsProvider extends OpenAICompatibleProvider<typeof provider.id> {
 	readonly id = provider.id;
@@ -63,11 +16,69 @@ export class GitHubModelsProvider extends OpenAICompatibleProvider<typeof provid
 		keyValidator: /(?:ghp-)?[a-zA-Z0-9]{32,}/,
 	};
 
-	getModels(): Promise<readonly AIModel<typeof provider.id>[]> {
-		return Promise.resolve(models);
+	async getModels(): Promise<readonly AIModel<typeof provider.id>[]> {
+		const rsp = await fetch('https://github.com/marketplace?category=All&task=chat-completion&type=models', {
+			headers: { Accept: 'application/json', 'Content-Type': 'application/json' },
+		});
+
+		interface ModelsResponseResult {
+			type: 'model';
+			task: 'chat-completion';
+
+			id: string;
+			name: string;
+			friendly_name: string;
+			publisher: string;
+			model_family: string;
+			max_input_tokens: number;
+			max_output_tokens: number;
+		}
+
+		interface ModelsResponse {
+			results: ModelsResponseResult[];
+		}
+
+		const result: ModelsResponse = await rsp.json();
+
+		const models = result.results.map(
+			r =>
+				({
+					id: r.name as any,
+					name: r.friendly_name,
+					maxTokens: { input: r.max_input_tokens, output: r.max_output_tokens },
+					provider: provider,
+					temperature: null,
+				}) satisfies GitHubModelsModel,
+		);
+
+		return models;
 	}
 
 	protected getUrl(_model: AIModel<typeof provider.id>): string {
 		return 'https://models.inference.ai.azure.com/chat/completions';
 	}
+
+	override async handleFetchFailure(
+		rsp: Response,
+		model: AIModel<typeof provider.id>,
+		retries: number,
+		maxCodeCharacters: number,
+	): Promise<{ retry: boolean; maxCodeCharacters: number }> {
+		if (rsp.status !== 404 && rsp.status !== 429) {
+			let json;
+			try {
+				json = (await rsp.json()) as { error?: { code: string; message: string } } | undefined;
+			} catch {}
+
+			if (retries < 2 && json?.error?.code === 'tokens_limit_reached') {
+				const match = /Max size: (\d+) tokens/.exec(json?.error?.message);
+				if (match?.[1] != null) {
+					maxCodeCharacters = getMaxCharacters(model, 2600, parseInt(match[1], 10));
+					return { retry: true, maxCodeCharacters: maxCodeCharacters };
+				}
+			}
+		}
+
+		return super.handleFetchFailure(rsp, model, retries, maxCodeCharacters);
+	}
 }

src/ai/huggingFaceProvider.ts

@@ -1,35 +1,10 @@
+import { fetch } from '@env/fetch';
 import type { AIModel } from './aiProviderService';
 import { OpenAICompatibleProvider } from './openAICompatibleProvider';
 
 const provider = { id: 'huggingface', name: 'Hugging Face' } as const;
 
 type HuggingFaceModel = AIModel<typeof provider.id>;
-const models: HuggingFaceModel[] = [
-	{
-		id: 'meta-llama/Llama-3.2-11B-Vision-Instruct',
-		name: 'Meta Llama 3.2 11B Vision',
-		maxTokens: { input: 131072, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'Qwen/Qwen2.5-72B-Instruct',
-		name: 'Qwen 2.5 72B',
-		maxTokens: { input: 131072, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'NousResearch/Hermes-3-Llama-3.1-8B',
-		name: 'Nous Research Hermes 3',
-		maxTokens: { input: 131072, output: 4096 },
-		provider: provider,
-	},
-	{
-		id: 'mistralai/Mistral-Nemo-Instruct-2407',
-		name: 'Mistral Nemo',
-		maxTokens: { input: 131072, output: 4096 },
-		provider: provider,
-	},
-];
 
 export class HuggingFaceProvider extends OpenAICompatibleProvider<typeof provider.id> {
 	readonly id = provider.id;
@@ -39,8 +14,44 @@ export class HuggingFaceProvider extends OpenAICompatibleProvider<typeof provide
 		keyValidator: /(?:hf_)?[a-zA-Z0-9]{32,}/,
 	};
 
-	getModels(): Promise<readonly AIModel<typeof provider.id>[]> {
-		return Promise.resolve(models);
+	async getModels(): Promise<readonly AIModel<typeof provider.id>[]> {
+		const apiKey = await this.getApiKey();
+
+		const query = new URLSearchParams({
+			filter: 'text-generation,conversational',
+			inference: 'warm',
+			sort: 'trendingScore',
+			limit: '30',
+		});
+		const rsp = await fetch(`https://huggingface.co/api/models?${query.toString()}`, {
+			headers: {
+				Accept: 'application/json',
+				'Content-Type': 'application/json',
+				Authorization: apiKey != null ? `Bearer ${apiKey}` : undefined!,
+			},
+			method: 'GET',
+		});
+
+		interface ModelsResponseResult {
+			id: string;
+		}
+
+		type ModelsResponse = ModelsResponseResult[];
+
+		const results: ModelsResponse = await rsp.json();
+
+		const models = results.map(
+			r =>
+				({
+					id: r.id,
+					name: r.id.split('/').pop()!,
+					maxTokens: { input: 4096, output: 4096 },
+					provider: provider,
+					temperature: null,
+				}) satisfies HuggingFaceModel,
+		);
+
+		return models;
 	}
 
 	protected getUrl(model: AIModel<typeof provider.id>): string {