// Original script was GPT4 but it has been deeply Ship of Theseused. import * as vscode from "vscode"; import axios from "axios"; let VSConfig: vscode.WorkspaceConfiguration; let apiEndpoint: string; let apiAuthentication: string; let apiModel: string; let apiMessageHeader: string; let apiTemperature: number; let numPredict: number; let promptWindowSize: number; let completionKeys: string; let responsePreview: boolean | undefined; let responsePreviewMaxTokens: number; let responsePreviewDelay: number; let continueInline: boolean | undefined; let keepAlive: number | undefined; let topK: number | undefined; let topP: number | undefined; let tfsZ: number | undefined; let typicalP: number | undefined; let repeatLastN: number | undefined; let repeatPenalty: number | undefined; let presencePenalty: number | undefined; let frequencyPenalty: number | undefined; let numBatch: number | undefined; let numKeep: number | undefined; function updateVSConfig() { VSConfig = vscode.workspace.getConfiguration("ollama-autocoder"); apiEndpoint = VSConfig.get("endpoint") || "http://localhost:11434/api/generate"; apiAuthentication = VSConfig.get("authentication") || ""; apiModel = VSConfig.get("model") || "openhermes2.5-mistral:7b-q4_K_M"; // The model I tested with apiMessageHeader = VSConfig.get("message header") || ""; numPredict = VSConfig.get("max tokens predicted") || 1000; promptWindowSize = VSConfig.get("prompt window size") || 2000; completionKeys = VSConfig.get("completion keys") || " "; responsePreview = VSConfig.get("response preview"); responsePreviewMaxTokens = VSConfig.get("preview max tokens") || 50; responsePreviewDelay = VSConfig.get("preview delay") || 0; // Must be || 0 instead of || [default] because of truthy continueInline = VSConfig.get("continue inline"); apiTemperature = VSConfig.get("temperature") || 0.5; keepAlive = VSConfig.get("keep alive") || undefined; topK = VSConfig.get("top k") || undefined; topP = VSConfig.get("top p") || undefined; tfsZ = VSConfig.get("tfs z") || undefined; typicalP = VSConfig.get("typical p") || undefined; repeatLastN = VSConfig.get("repeat last n") || undefined; repeatPenalty = VSConfig.get("repeat penalty") || undefined; presencePenalty = VSConfig.get("presence penalty") || undefined; frequencyPenalty = VSConfig.get("frequency penalty") || undefined; numBatch = VSConfig.get("num batch") || undefined; numKeep = VSConfig.get("num keep") || undefined; } updateVSConfig(); // No need for restart for any of these settings vscode.workspace.onDidChangeConfiguration(updateVSConfig); // Give model additional information function messageHeaderSub(document: vscode.TextDocument) { const sub = apiMessageHeader .replace("{LANG}", document.languageId) .replace("{FILE_NAME}", document.fileName) .replace("{PROJECT_NAME}", vscode.workspace.name || "Untitled"); return sub; } // internal function for autocomplete, not directly exposed async function autocompleteCommand(textEditor: vscode.TextEditor, cancellationToken?: vscode.CancellationToken) { const document = textEditor.document; const position = textEditor.selection.active; // Get the current prompt let prompt = document.getText(new vscode.Range(document.lineAt(0).range.start, position)); prompt = prompt.substring(Math.max(0, prompt.length - promptWindowSize), prompt.length); // Show a progress message vscode.window.withProgress( { location: vscode.ProgressLocation.Notification, title: "Ollama Autocoder", cancellable: true, }, async (progress, progressCancellationToken) => { try { progress.report({ message: "Starting model..." }); let axiosCancelPost: () => void; const axiosCancelToken = new axios.CancelToken((c) => { const cancelPost = function () { c("Autocompletion request terminated by user cancel"); }; axiosCancelPost = cancelPost; if (cancellationToken) cancellationToken.onCancellationRequested(cancelPost); progressCancellationToken.onCancellationRequested(cancelPost); vscode.workspace.onDidCloseTextDocument(cancelPost); }); // Make a request to the ollama.ai REST API const response = await axios.post(apiEndpoint, { model: apiModel, // Change this to the model you want to use prompt: messageHeaderSub(textEditor.document) + prompt, stream: true, raw: true, options: { num_predict: numPredict, temperature: apiTemperature, stop: ["```"] } }, { cancelToken: axiosCancelToken, responseType: 'stream', headers: { 'Authorization': apiAuthentication } } ); //tracker let currentPosition = position; response.data.on('data', async (d: Uint8Array) => { progress.report({ message: "Generating..." }); // Check for user input (cancel) if (currentPosition.line != textEditor.selection.end.line || currentPosition.character != textEditor.selection.end.character) { axiosCancelPost(); // cancel axios => cancel finished promise => close notification return; } // Get a completion from the response const completion: string = JSON.parse(d.toString()).response; // lastToken = completion; if (completion === "") { return; } //complete edit for token const edit = new vscode.WorkspaceEdit(); edit.insert(document.uri, currentPosition, completion); await vscode.workspace.applyEdit(edit); // Move the cursor to the end of the completion const completionLines = completion.split("\n"); const newPosition = new vscode.Position( currentPosition.line + completionLines.length - 1, (completionLines.length > 1 ? 0 : currentPosition.character) + completionLines[completionLines.length - 1].length ); const newSelection = new vscode.Selection( position, newPosition ); currentPosition = newPosition; // completion bar progress.report({ message: "Generating...", increment: 1 / (numPredict / 100) }); // move cursor textEditor.selection = newSelection; }); // Keep cancel window available const finished = new Promise((resolve) => { response.data.on('end', () => { progress.report({ message: "Ollama completion finished." }); resolve(true); }); axiosCancelToken.promise.finally(() => { // prevent notification from freezing on user input cancel resolve(false); }); }); await finished; } catch (err: any) { // Show an error message vscode.window.showErrorMessage( "Ollama encountered an error: " + err.message ); console.log(err); } } ); } // Completion item provider callback for activate async function provideCompletionItems(document: vscode.TextDocument, position: vscode.Position, cancellationToken: vscode.CancellationToken) { // Create a completion item const item = new vscode.CompletionItem("Autocomplete with Ollama"); // Set the insert text to a placeholder item.insertText = new vscode.SnippetString('${1:}'); // Wait before initializing Ollama to reduce compute usage if (responsePreview) await new Promise(resolve => setTimeout(resolve, responsePreviewDelay * 1000)); if (cancellationToken.isCancellationRequested) { return [ item ]; } // Set the label & inset text to a shortened, non-stream response if (responsePreview) { let prompt = document.getText(new vscode.Range(document.lineAt(0).range.start, position)); prompt = prompt.substring(Math.max(0, prompt.length - promptWindowSize), prompt.length); const response_preview = await axios.post(apiEndpoint, { model: apiModel, // Change this to the model you want to use prompt: messageHeaderSub(document) + prompt, stream: false, raw: true, options: { num_predict: responsePreviewMaxTokens, // reduced compute max temperature: apiTemperature, stop: ['\n', '```'], ...keepAlive && { keep_alive: keepAlive }, ...topK && { top_k: topK }, ...topP && { top_p: topP }, ...tfsZ && { tfs_z: tfsZ }, ...typicalP && { typical_p: typicalP }, ...repeatLastN && { repeat_last_n: repeatLastN }, ...repeatPenalty && { repeat_penalty: repeatPenalty }, ...presencePenalty && { presence_penalty: presencePenalty }, ...frequencyPenalty && { frequency_penalty: frequencyPenalty }, ...numBatch && { num_batch: numBatch }, ...numKeep && { num_keep: numKeep }, } }, { cancelToken: new axios.CancelToken((c) => { const cancelPost = function () { c("Autocompletion request terminated by completion cancel"); }; cancellationToken.onCancellationRequested(cancelPost); }) }); if (response_preview.data.response.trim() != "") { // default if empty item.label = response_preview.data.response.trimStart(); // tended to add whitespace at the beginning item.insertText = response_preview.data.response.trimStart(); } } // Set the documentation to a message item.documentation = new vscode.MarkdownString('Press `Enter` to get an autocompletion from Ollama'); // Set the command to trigger the completion if (continueInline || !responsePreview) item.command = { command: 'ollama-autocoder.autocomplete', title: 'Autocomplete with Ollama', arguments: [cancellationToken] }; // Return the completion item return [item]; } // This method is called when extension is activated function activate(context: vscode.ExtensionContext) { // Register a completion provider for JavaScript files const completionProvider = vscode.languages.registerCompletionItemProvider("*", { provideCompletionItems }, ...completionKeys.split("") ); // Register a command for getting a completion from Ollama through command/keybind const externalAutocompleteCommand = vscode.commands.registerTextEditorCommand( "ollama-autocoder.autocomplete", (textEditor, _, cancellationToken?) => { // no cancellation token from here, but there is one from completionProvider autocompleteCommand(textEditor, cancellationToken); } ); // Add the commands & completion provider to the context context.subscriptions.push(completionProvider); context.subscriptions.push(externalAutocompleteCommand); } // This method is called when extension is deactivated // eslint-disable-next-line @typescript-eslint/no-empty-function function deactivate() { } module.exports = { activate, deactivate, };