autocomplete draft

bjoaquinc · Jan 28, 2025 · e4bb15e · e4bb15e
1 parent e4d747d
commit e4bb15e
Show file tree

Hide file tree

Showing 6 changed files with 217 additions and 86 deletions.
diff --git a/src/vs/platform/void/common/llmMessageTypes.ts b/src/vs/platform/void/common/llmMessageTypes.ts
@@ -34,16 +34,18 @@ export type _InternalLLMMessage = {
 	content: string;
 }
 
+type _InternalOllamaFIMMessages = {
+	prefix: string;
+	suffix: string;
+	stopTokens: string[];
+}
 
 type SendLLMType = {
 	type: 'sendLLMMessage';
 	messages: LLMMessage[];
 } | {
 	type: 'ollamaFIM';
-	messages: {
-		prefix: string;
-		suffix: string;
-	}
+	messages: _InternalOllamaFIMMessages;
 }
 
 // service types
@@ -56,7 +58,7 @@ export type ServiceSendLLMMessageParams = {
 } & SendLLMType
 
 // params to the true sendLLMMessage function
-export type SendLLMMMessageParams = {
+export type SendLLMMessageParams = {
 	onText: OnText;
 	onFinalMessage: OnFinalMessage;
 	onError: OnError;
@@ -74,26 +76,40 @@ export type SendLLMMMessageParams = {
 
 // can't send functions across a proxy, use listeners instead
 export type BlockedMainLLMMessageParams = 'onText' | 'onFinalMessage' | 'onError' | 'abortRef'
-export type MainSendLLMMessageParams = Omit<SendLLMMMessageParams, BlockedMainLLMMessageParams> & { requestId: string } & SendLLMType
+export type MainSendLLMMessageParams = Omit<SendLLMMessageParams, BlockedMainLLMMessageParams> & { requestId: string } & SendLLMType
 
 export type MainLLMMessageAbortParams = { requestId: string }
 
 export type EventLLMMessageOnTextParams = Parameters<OnText>[0] & { requestId: string }
 export type EventLLMMessageOnFinalMessageParams = Parameters<OnFinalMessage>[0] & { requestId: string }
 export type EventLLMMessageOnErrorParams = Parameters<OnError>[0] & { requestId: string }
 
+
 export type _InternalSendLLMMessageFnType = (
 	params: {
 		onText: OnText;
 		onFinalMessage: OnFinalMessage;
 		onError: OnError;
+		providerName: ProviderName;
+		settingsOfProvider: SettingsOfProvider;
+		modelName: string;
+		_setAborter: (aborter: () => void) => void;
+
 		messages: _InternalLLMMessage[];
+	}
+) => void
 
-		settingsOfProvider: SettingsOfProvider;
+export type _InternalOllamaFIMMessageFnType = (
+	params: {
+		onText: OnText;
+		onFinalMessage: OnFinalMessage;
+		onError: OnError;
 		providerName: ProviderName;
+		settingsOfProvider: SettingsOfProvider;
 		modelName: string;
-
 		_setAborter: (aborter: () => void) => void;
+
+		messages: _InternalOllamaFIMMessages;
 	}
 ) => void
 

diff --git a/src/vs/platform/void/electron-main/llmMessage/ollama.ts b/src/vs/platform/void/electron-main/llmMessage/ollama.ts
@@ -4,7 +4,7 @@
  *--------------------------------------------------------------------------------------*/
 
 import { Ollama } from 'ollama';
-import { _InternalModelListFnType, _InternalSendLLMMessageFnType, OllamaModelResponse } from '../../common/llmMessageTypes.js';
+import { _InternalModelListFnType, _InternalOllamaFIMMessageFnType, _InternalSendLLMMessageFnType, OllamaModelResponse } from '../../common/llmMessageTypes.js';
 import { defaultProviderSettings } from '../../common/voidSettingsTypes.js';
 
 export const ollamaList: _InternalModelListFnType<OllamaModelResponse> = async ({ onSuccess: onSuccess_, onError: onError_, settingsOfProvider }) => {
@@ -25,7 +25,7 @@ export const ollamaList: _InternalModelListFnType<OllamaModelResponse> = async (
 		const ollama = new Ollama({ host: thisConfig.endpoint })
 		ollama.list()
 			.then((response) => {
-				console.log('MODELS!!!!!!!!!!!!!!!!!', response)
+				// console.log('MODELS!!!!!!!!!!!!!!!!!', response)
 				const { models } = response
 				onSuccess({ models })
 			})
@@ -39,6 +39,44 @@ export const ollamaList: _InternalModelListFnType<OllamaModelResponse> = async (
 }
 
 
+export const sendOllamaFIM: _InternalOllamaFIMMessageFnType = ({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter }) => {
+
+	const thisConfig = settingsOfProvider.ollama
+	// if endpoint is empty, normally ollama will send to 11434, but we want it to fail - the user should type it in
+	if (!thisConfig.endpoint) throw new Error(`Ollama Endpoint was empty (please enter ${defaultProviderSettings.ollama.endpoint} if you want the default).`)
+
+	let fullText = ''
+
+	const ollama = new Ollama({ host: thisConfig.endpoint })
+
+	ollama.generate({
+		model: modelName,
+		prompt: messages.prefix,
+		suffix: messages.suffix,
+		options: {
+			stop: messages.stopTokens,
+		},
+		raw: true,
+		stream: true,
+		// options: { num_predict: parseMaxTokensStr(thisConfig.maxTokens) } // this is max_tokens
+	})
+		.then(async stream => {
+			_setAborter(() => stream.abort())
+			// iterate through the stream
+			for await (const chunk of stream) {
+				const newText = chunk.response;
+				fullText += newText;
+				onText({ newText, fullText });
+			}
+			onFinalMessage({ fullText });
+			console.log('!!!!! OLLAMA RESULT', JSON.stringify(fullText))
+		})
+		// when error/fail
+		.catch((error) => {
+			onError({ message: error + '', fullError: error })
+		})
+};
+
 
 // Ollama
 export const sendOllamaMsg: _InternalSendLLMMessageFnType = ({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter }) => {

diff --git a/src/vs/platform/void/electron-main/llmMessage/sendLLMMessage.ts b/src/vs/platform/void/electron-main/llmMessage/sendLLMMessage.ts
@@ -3,11 +3,11 @@
  *  Licensed under the Apache License, Version 2.0. See LICENSE.txt for more information.
  *--------------------------------------------------------------------------------------*/
 
-import { SendLLMMMessageParams, OnText, OnFinalMessage, OnError, LLMMessage, _InternalLLMMessage } from '../../common/llmMessageTypes.js';
+import { SendLLMMessageParams, OnText, OnFinalMessage, OnError, LLMMessage, _InternalLLMMessage } from '../../common/llmMessageTypes.js';
 import { IMetricsService } from '../../common/metricsService.js';
 
 import { sendAnthropicMsg } from './anthropic.js';
-import { sendOllamaMsg } from './ollama.js';
+import { sendOllamaFIM, sendOllamaMsg } from './ollama.js';
 import { sendOpenAIMsg } from './openai.js';
 import { sendGeminiMsg } from './gemini.js';
 import { sendGroqMsg } from './groq.js';
@@ -59,25 +59,22 @@ export const sendLLMMessage = ({
 	settingsOfProvider,
 	providerName,
 	modelName,
-}: SendLLMMMessageParams,
+}: SendLLMMessageParams,
 
 	metricsService: IMetricsService
 ) => {
-	messages.unshift({ role: 'system', content: aiInstructions })
+	// messages.unshift({ role: 'system', content: aiInstructions })
 
-	const messages = type === 'sendLLMMessage' ? cleanMessages(messages_) : []
-
-
-	const prefixAndSuffix = type === 'ollamaFIM' ? messages_ : null
+	const messagesArr = type === 'sendLLMMessage' ? cleanMessages(messages_) : []
 
 	// only captures number of messages and message "shape", no actual code, instructions, prompts, etc
 	const captureLLMEvent = (eventId: string, extras?: object) => {
 		metricsService.capture(eventId, {
 			providerName,
 			modelName,
 			...type === 'sendLLMMessage' ? {
-				numMessages: messages?.length,
-				messagesShape: messages?.map(msg => ({ role: msg.role, length: msg.content.length })),
+				numMessages: messagesArr?.length,
+				messagesShape: messagesArr?.map(msg => ({ role: msg.role, length: msg.content.length })),
 				origNumMessages: messages_?.length,
 				origMessagesShape: messages_?.map(msg => ({ role: msg.role, length: msg.content.length })),
 
@@ -122,27 +119,30 @@ export const sendLLMMessage = ({
 	}
 	abortRef_.current = onAbort
 
-	captureLLMEvent(`${loggingName} - Sending Message`, { messageLength: messages[messages.length - 1]?.content.length })
+	captureLLMEvent(`${loggingName} - Sending Message`, { messageLength: messagesArr[messagesArr.length - 1]?.content.length })
 
 	try {
 		switch (providerName) {
 			case 'anthropic':
-				sendAnthropicMsg({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
+				sendAnthropicMsg({ messages: messagesArr, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
 				break;
 			case 'openAI':
 			case 'openRouter':
 			case 'deepseek':
 			case 'openAICompatible':
-				sendOpenAIMsg({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
+				sendOpenAIMsg({ messages: messagesArr, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
 				break;
 			case 'gemini':
-				sendGeminiMsg({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
+				sendGeminiMsg({ messages: messagesArr, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
 				break;
 			case 'ollama':
-				sendOllamaMsg({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
+				if (type === 'ollamaFIM')
+					sendOllamaFIM({ messages: messages_, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName })
+				else
+					sendOllamaMsg({ messages: messagesArr, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
 				break;
 			case 'groq':
-				sendGroqMsg({ messages, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
+				sendGroqMsg({ messages: messagesArr, onText, onFinalMessage, onError, settingsOfProvider, modelName, _setAborter, providerName });
 				break;
 			default:
 				onError({ message: `Error: Void provider was "${providerName}", which is not recognized.`, fullError: null })

diff --git a/src/vs/platform/void/electron-main/llmMessageChannel.ts b/src/vs/platform/void/electron-main/llmMessageChannel.ts
@@ -8,7 +8,7 @@
 
 import { IServerChannel } from '../../../base/parts/ipc/common/ipc.js';
 import { Emitter, Event } from '../../../base/common/event.js';
-import { EventLLMMessageOnTextParams, EventLLMMessageOnErrorParams, EventLLMMessageOnFinalMessageParams, MainSendLLMMessageParams, AbortRef, SendLLMMMessageParams, MainLLMMessageAbortParams, MainModelListParams, ModelListParams, EventModelListOnSuccessParams, EventModelListOnErrorParams, OllamaModelResponse, OpenaiCompatibleModelResponse, } from '../common/llmMessageTypes.js';
+import { EventLLMMessageOnTextParams, EventLLMMessageOnErrorParams, EventLLMMessageOnFinalMessageParams, MainSendLLMMessageParams, AbortRef, SendLLMMessageParams, MainLLMMessageAbortParams, MainModelListParams, ModelListParams, EventModelListOnSuccessParams, EventModelListOnErrorParams, OllamaModelResponse, OpenaiCompatibleModelResponse, } from '../common/llmMessageTypes.js';
 import { sendLLMMessage } from './llmMessage/sendLLMMessage.js'
 import { IMetricsService } from '../common/metricsService.js';
 import { ollamaList } from './llmMessage/ollama.js';
@@ -97,7 +97,7 @@ export class LLMMessageChannel implements IServerChannel {
 		if (!(requestId in this._abortRefOfRequestId_llm))
 			this._abortRefOfRequestId_llm[requestId] = { current: null }
 
-		const mainThreadParams: SendLLMMMessageParams = {
+		const mainThreadParams: SendLLMMessageParams = {
 			...params,
 			onText: ({ newText, fullText }) => { this._onText_llm.fire({ requestId, newText, fullText }); },
 			onFinalMessage: ({ fullText }) => { this._onFinalMessage_llm.fire({ requestId, fullText }); },