Fixes #4071 adapts openai compat for anthropic

Ensures gemini still includes max_tokens

Author: eamodio

CHANGELOG.md

@@ -8,6 +8,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/) and this p
 
 ### Fixed
 
+- Fixes Generate Commit Message Error - Anthropic ([#4071](https://github.com/gitkraken/vscode-gitlens/issues/4071))
 - Fixes Settings editor breaking when dragging it to a new tab group ([#4061](https://github.com/gitkraken/vscode-gitlens/issues/4061))
 - Fixes regression where hovering over the Graph WIP row doesn't show up anymore ([#4062](https://github.com/gitkraken/vscode-gitlens/issues/4062))
 - Fixes Inspect & Graph Details: autolinks rendering when enabled setting is false ([#3841](https://github.com/gitkraken/vscode-gitlens/issues/3841))

src/ai/anthropicProvider.ts

@@ -1,7 +1,6 @@
 import type { CancellationToken } from 'vscode';
+import type { Response } from '@env/fetch';
 import type { AIModel } from './aiProviderService';
-import { getMaxCharacters, getValidatedTemperature } from './aiProviderService';
-import type { ChatMessage } from './openAICompatibleProvider';
 import { OpenAICompatibleProvider } from './openAICompatibleProvider';
 
 const provider = { id: 'anthropic', name: 'Anthropic' } as const;
@@ -98,66 +97,50 @@ export class AnthropicProvider extends OpenAICompatibleProvider<typeof provider.
 		};
 	}
 
-	override async fetch(
+	protected override fetchCore(
 		model: AIModel<typeof provider.id>,
 		apiKey: string,
-		messages: (maxCodeCharacters: number, retries: number) => ChatMessage[],
-		outputTokens: number,
+		request: object,
 		cancellation: CancellationToken | undefined,
-	): Promise<[result: string, maxCodeCharacters: number]> {
-		let retries = 0;
-		let maxCodeCharacters = getMaxCharacters(model, 2600);
-
-		while (true) {
-			// Split the system message from the rest of the messages
-			const [system, ...msgs] = messages(maxCodeCharacters, retries);
-
-			const request: AnthropicMessageRequest = {
-				model: model.id,
-				messages: msgs,
-				system: system.content,
-				stream: false,
-				max_tokens: Math.min(outputTokens, model.maxTokens.output),
-				temperature: getValidatedTemperature(model.temperature),
-			};
-
-			const rsp = await this.fetchCore(model, apiKey, request, cancellation);
-			if (!rsp.ok) {
-				if (rsp.status === 404) {
-					throw new Error(`Your API key doesn't seem to have access to the selected '${model.id}' model`);
-				}
-				if (rsp.status === 429) {
-					throw new Error(
-						`(${this.name}) ${rsp.status}: Too many requests (rate limit exceeded) or your API key is associated with an expired trial`,
-					);
-				}
-
-				let json;
-				try {
-					json = (await rsp.json()) as AnthropicError | undefined;
-				} catch {}
+	): Promise<Response> {
+		if ('max_completion_tokens' in request) {
+			const { max_completion_tokens: max, ...rest } = request;
+			request = max ? { max_tokens: max, ...rest } : rest;
+		}
+		return super.fetchCore(model, apiKey, request, cancellation);
+	}
 
-				debugger;
+	protected override async handleFetchFailure(
+		rsp: Response,
+		model: AIModel<typeof provider.id>,
+		retries: number,
+		maxCodeCharacters: number,
+	): Promise<{ retry: boolean; maxCodeCharacters: number }> {
+		if (rsp.status === 404) {
+			throw new Error(`Your API key doesn't seem to have access to the selected '${model.id}' model`);
+		}
+		if (rsp.status === 429) {
+			throw new Error(
+				`(${this.name}) ${rsp.status}: Too many requests (rate limit exceeded) or your account is out of funds`,
+			);
+		}
 
-				if (
-					retries++ < 2 &&
-					json?.error?.type === 'invalid_request_error' &&
-					json?.error?.message?.includes('prompt is too long')
-				) {
-					maxCodeCharacters -= 500 * retries;
-					continue;
-				}
+		let json;
+		try {
+			json = (await rsp.json()) as AnthropicError | undefined;
+		} catch {}
 
-				throw new Error(`(${this.name}) ${rsp.status}: ${json?.error?.message || rsp.statusText})`);
-			}
+		debugger;
 
-			const data: AnthropicMessageResponse = await rsp.json();
-			const result = data.content
-				.map(c => c.text)
-				.join('\n')
-				.trim();
-			return [result, maxCodeCharacters];
+		if (
+			retries++ < 2 &&
+			json?.error?.type === 'invalid_request_error' &&
+			json?.error?.message?.includes('prompt is too long')
+		) {
+			return { retry: true, maxCodeCharacters: maxCodeCharacters - 500 * retries };
 		}
+
+		throw new Error(`(${this.name}) ${rsp.status}: ${json?.error?.message || rsp.statusText})`);
 	}
 }
 
@@ -175,31 +158,3 @@ interface AnthropicError {
 		message: string;
 	};
 }
-
-interface AnthropicMessageRequest {
-	model: AnthropicModel['id'];
-	messages: ChatMessage[];
-	system?: string;
-
-	max_tokens: number;
-	metadata?: object;
-	stop_sequences?: string[];
-	stream?: boolean;
-	temperature?: number;
-	top_p?: number;
-	top_k?: number;
-}
-
-interface AnthropicMessageResponse {
-	id: string;
-	type: 'message';
-	role: 'assistant';
-	content: { type: 'text'; text: string }[];
-	model: string;
-	stop_reason: 'end_turn' | 'max_tokens' | 'stop_sequence';
-	stop_sequence: string | null;
-	usage: {
-		input_tokens: number;
-		output_tokens: number;
-	};
-}

src/ai/geminiProvider.ts

@@ -99,8 +99,8 @@ export class GeminiProvider extends OpenAICompatibleProvider<typeof provider.id>
 		cancellation: CancellationToken | undefined,
 	): Promise<Response> {
 		if ('max_completion_tokens' in request) {
-			const { max_completion_tokens: _, ...rest } = request;
-			request = rest;
+			const { max_completion_tokens: max, ...rest } = request;
+			request = max ? { max_tokens: max, ...rest } : rest;
 		}
 		return super.fetchCore(model, apiKey, request, cancellation);
 	}

src/ai/openAICompatibleProvider.ts

@@ -256,7 +256,7 @@ export abstract class OpenAICompatibleProvider<T extends AIProviders> implements
 			}
 
 			const data: ChatCompletionResponse = await rsp.json();
-			const result = data.choices[0].message.content?.trim() ?? '';
+			const result = data.choices?.[0].message.content?.trim() ?? data.content?.[0]?.text?.trim() ?? '';
 			return [result, maxCodeCharacters];
 		}
 	}
@@ -331,24 +331,23 @@ interface ChatCompletionRequest {
 	model: string;
 	messages: ChatMessage<Role>[];
 
-	frequency_penalty?: number;
-	logit_bias?: Record<string, number>;
+	/** @deprecated */
+	max_tokens?: number;
 	max_completion_tokens?: number;
-	n?: number;
-	presence_penalty?: number;
-	stop?: string | string[];
+	metadata?: Record<string, string>;
 	stream?: boolean;
 	temperature?: number;
 	top_p?: number;
-	user?: string;
+
+	/** Not supported by many models/providers */
+	reasoning_effort?: 'low' | 'medium' | 'high';
 }
 
 interface ChatCompletionResponse {
 	id: string;
-	object: 'chat.completion';
-	created: number;
 	model: string;
-	choices: {
+	/** OpenAI compatible output */
+	choices?: {
 		index: number;
 		message: {
 			role: Role;
@@ -357,9 +356,15 @@ interface ChatCompletionResponse {
 		};
 		finish_reason: string;
 	}[];
+	/** Anthropic compatible output */
+	content?: { type: 'text'; text: string }[];
 	usage: {
-		prompt_tokens: number;
-		completion_tokens: number;
-		total_tokens: number;
+		input_tokens?: number;
+		prompt_tokens?: number;
+
+		completion_tokens?: number;
+		output_tokens?: number;
+
+		total_tokens?: number;
 	};
 }