misc

src/vs/platform/void/common/toolsService.ts

@@ -25,7 +25,7 @@ export type InternalToolInfo = {
 // helper
 const pagination = {
 	desc: `Very large results may be paginated (indicated in the result). Pagination fails gracefully if out of bounds or invalid page number.`,
-	param: { pageNumber: { type: 'number', description: 'The page number (optional, defaults to 1).' }, }
+	param: { pageNumber: { type: 'number', description: 'The page number (optional, default is 1).' }, }
 } as const
 
 export const contextTools = {
@@ -71,7 +71,7 @@ export const contextTools = {
 
 } as const satisfies { [name: string]: InternalToolInfo }
 
-type ContextToolName = keyof typeof contextTools
+export type ContextToolName = keyof typeof contextTools
 type ContextToolParamNames<T extends ContextToolName> = keyof typeof contextTools[T]['params']
 type ContextToolParams<T extends ContextToolName> = { [paramName in ContextToolParamNames<T>]: unknown }
 

src/vs/platform/void/electron-main/llmMessage/groq.ts

@@ -22,8 +22,6 @@ export const sendGroqChat: _InternalSendLLMChatMessageFnType = async ({ messages
 			messages: messages,
 			model: modelName,
 			stream: true,
-			// temperature: 0.7,
-			// max_tokens: parseMaxTokensStr(thisConfig.maxTokens),
 		})
 		.then(async response => {
 			_setAborter(() => response.controller.abort())

src/vs/platform/void/electron-main/llmMessage/mistral.ts

@@ -21,8 +21,6 @@ export const sendMistralChat: _InternalSendLLMChatMessageFnType = async ({ messa
 			messages: messages,
 			model: modelName,
 			stream: true,
-			// temperature: 0.7,
-			// maxTokens: 2048,
 		})
 		.then(async response => {
 			// Mistral has a really nonstandard API - no interrupt and weird stream types

src/vs/platform/void/electron-main/llmMessage/openai.ts

@@ -6,7 +6,7 @@
 import OpenAI from 'openai';
 import { _InternalModelListFnType, _InternalSendLLMFIMMessageFnType, _InternalSendLLMChatMessageFnType } from '../../common/llmMessageTypes.js';
 import { Model } from 'openai/resources/models.js';
-import { contextTools, InternalToolInfo } from '../../common/toolsService.js';
+import { InternalToolInfo } from '../../common/toolsService.js';
 // import { parseMaxTokensStr } from './util.js';
 
 
@@ -141,7 +141,7 @@ export const sendOpenAIChat: _InternalSendLLMChatMessageFnType = ({ messages, on
 		model: modelName,
 		messages: messages,
 		stream: true,
-		tools: [toOpenAITool('list_dir', contextTools['list_dir'])],
+		// tools: Object.keys(contextTools).map(name => toOpenAITool(name, contextTools[name as ContextToolName])),
 	}
 
 	openai.chat.completions
@@ -150,8 +150,6 @@ export const sendOpenAIChat: _InternalSendLLMChatMessageFnType = ({ messages, on
 			_setAborter(() => response.controller.abort())
 			// when receive text
 			for await (const chunk of response) {
-				console.log('!!!', JSON.stringify(chunk, null, 2))
-
 
 				let newText = ''
 				newText += chunk.choices[0]?.delta?.tool_calls?.[0]?.function?.name ?? ''

src/vs/platform/void/electron-main/llmMessage/sendLLMMessage.ts

@@ -65,9 +65,14 @@ export const sendLLMMessage = ({
 
 	metricsService: IMetricsService
 ) => {
-	// messages.unshift({ role: 'system', content: aiInstructions })
 
-	const messagesArr = messagesType === 'chatMessages' ? cleanChatMessages(messages_) : []
+	let messagesArr: _InternalLLMChatMessage[] = []
+	if (messagesType === 'chatMessages') {
+		messagesArr = cleanChatMessages([
+			{ role: 'system', content: aiInstructions },
+			...messages_
+		])
+	}
 
 	// only captures number of messages and message "shape", no actual code, instructions, prompts, etc
 	const captureLLMEvent = (eventId: string, extras?: object) => {

src/vs/platform/void/electron-main/templates/templates.ts

@@ -5,9 +5,8 @@ modelName -> {
 	supports_tools: boolean // we will just do a string of tool use if it doesn't support
 	supports_autocomplete_FIM (suffix) // we will just do a description of FIM if it doens't support <|fim_hole|>
 
-	supports_streaming: boolean (o1 does NOT)
-	max_tokens: number
-
+	supports_streaming: boolean // (o1 does NOT) we will just dump the final result if doesn't support it
+	max_tokens: number // required, DEFAULT is Infinity
 
 }