lemonade-interactive-loader 1.0.3 → 1.0.4

package/README.md

@@ -83,16 +83,17 @@ The main menu adapts based on whether you have a configuration saved:
 
 ### The Setup Wizard
 
-Just answer 8 simple questions:
+Just answer 9 simple questions:
 
 1. **Network access?** Should the server be accessible from other devices?
 2. **Port number?** Which port should it run on? (default: 8080)
 3. **Logging level?** Choose from info, debug, warning, or error
-4. **Model directory?** Point to existing models (like LM Studio) if needed
-5. **Interface type?** System tray or headless mode
-6. **Custom arguments?** Any additional llama.cpp parameters?
-7. **Custom build?** Use a specific llama.cpp build from GitHub?
-8. **Backend?** Choose auto, vulkan, rocm, or cpu
+4. **Context window size?** Choose from 4K, 8K, 16K, 32K, 64K, 128K, or 256K tokens (default: 4K)
+5. **Model directory?** Point to existing models (like LM Studio) if needed
+6. **Interface type?** System tray or headless mode
+7. **Custom arguments?** Any additional llama.cpp parameters?
+8. **Custom build?** Use a specific llama.cpp build from GitHub?
+9. **Backend?** Choose auto, vulkan, rocm, or cpu
 
 ## ✨ Key Features
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lemonade-interactive-loader",
-  "version": "1.0.3",
+  "version": "1.0.4",
   "description": "Interactive CLI tool to launch Lemonade Server with custom arguments and download llama.cpp releases - Cross-platform (Windows/Linux)",
   "main": "index.js",
   "bin": {
@@ -35,6 +35,7 @@
     "adm-zip": "^0.5.16",
     "inquirer": "^8.2.6",
     "tar": "^7.5.9",
+    "tree-kill": "^1.2.2",
     "unzipper": "^0.12.3"
   }
 }

package/src/README.md

@@ -43,7 +43,7 @@ src/
 ### CLI (`cli/`)
 - **menu.js**: Main menu system and command routing
 - **prompts.js**: Interactive prompts for release/asset selection
-- **setup-wizard.js**: 8-question setup wizard
+- **setup-wizard.js**: 9-question setup wizard
 
 ## Architecture Principles
 

package/src/cli/menu.js

@@ -213,6 +213,7 @@ async function viewConfiguration() {
   console.log(`Host: ${config.host}`);
   console.log(`Port: ${config.port}`);
   console.log(`Log Level: ${config.logLevel}`);
+  console.log(`Context Size: ${config.contextSize}`);
   console.log(`Backend: ${config.backend}`);
   console.log(`Model Directory: ${config.modelDir}`);
   console.log(`Run Mode: ${config.runMode}`);

package/src/cli/prompts.js

@@ -223,6 +223,7 @@ function displayConfigSummary(config) {
   console.log(`Host: ${config.host}`);
   console.log(`Port: ${config.port}`);
   console.log(`Log Level: ${config.logLevel}`);
+  console.log(`Context Size: ${config.contextSize}`);
   console.log(`Backend: ${config.backend}`);
   console.log(`Model Directory: ${config.modelDir}`);
   console.log(`Run Mode: ${config.runMode}`);

package/src/cli/setup-wizard.js

@@ -1,5 +1,5 @@
 const inquirer = require('inquirer');
-const { DEFAULTS, BACKEND_TYPES, LOG_LEVELS, RUN_MODES, HOSTS } = require('../config/constants');
+const { DEFAULTS, BACKEND_TYPES, LOG_LEVELS, RUN_MODES, HOSTS, CONTEXT_SIZES } = require('../config/constants');
 const { loadConfig, saveConfig } = require('../config');
 const { selectLlamaCppRelease, selectAsset, selectInstalledAssetPrompt, displayConfigSummary } = require('./prompts');
 const { downloadAndExtractLlamaCpp } = require('../services/asset-manager');
@@ -54,9 +54,35 @@ async function runSetupWizard(isEdit = false) {
     }
   ]);
   
-  // Q4: Custom model directory
+  // Q4: Context window size
+  const savedContextSize = existingConfig.contextSize || DEFAULTS.CONTEXT_SIZE;
+  const contextSizeChoices = [
+    { name: '4K (4096 tokens) - Default', value: CONTEXT_SIZES['4K'] },
+    { name: '8K (8192 tokens)', value: CONTEXT_SIZES['8K'] },
+    { name: '16K (16384 tokens)', value: CONTEXT_SIZES['16K'] },
+    { name: '32K (32768 tokens)', value: CONTEXT_SIZES['32K'] },
+    { name: '64K (65536 tokens)', value: CONTEXT_SIZES['64K'] },
+    { name: '128K (131072 tokens)', value: CONTEXT_SIZES['128K'] },
+    { name: '256K (262144 tokens)', value: CONTEXT_SIZES['256K'] }
+  ];
+  
+  // Find the index of the saved/default context size
+  const defaultContextSizeIndex = contextSizeChoices.findIndex(choice => choice.value === savedContextSize);
+  const defaultContextSize = defaultContextSizeIndex >= 0 ? defaultContextSizeIndex : 0;
+  
+  const { contextSize } = await inquirer.prompt([
+    {
+      type: 'list',
+      name: 'contextSize',
+      message: 'How big should the context window be?',
+      choices: contextSizeChoices,
+      default: defaultContextSize
+    }
+  ]);
+  
+  // Q5: Custom model directory
   const existingModelDir = existingConfig.modelDir;
-  const hasExistingModelDir = existingModelDir && existingModelDir !== 'None';
+  const hasExistingModelDir = existingModelDir !== undefined;
   
   const { useCustomModelDir } = await inquirer.prompt([
     {
@@ -83,7 +109,7 @@ async function runSetupWizard(isEdit = false) {
     finalModelDir = DEFAULTS.MODEL_DIR;
   }
   
-  // Q5: System tray vs headless
+  // Q6: System tray vs headless
   const { runMode } = await inquirer.prompt([
     {
       type: 'list',
@@ -97,7 +123,7 @@ async function runSetupWizard(isEdit = false) {
     }
   ]);
   
-  // Q6: Custom llama.cpp args
+  // Q7: Custom llama.cpp args
   const existingLlamacppArgs = existingConfig.llamacppArgs || '';
   const hasExistingArgs = existingLlamacppArgs.length > 0;
   
@@ -117,7 +143,7 @@ async function runSetupWizard(isEdit = false) {
       {
         type: 'input',
         name: 'llamacppArgs',
-        message: 'Enter llama.cpp arguments (comma-separated, e.g., --ctx-size 4096,--batch-size 512):',
+        message: 'Enter llama.cpp arguments (comma-separated, e.g., --no-mmap,--batch-size 512):',
         default: existingLlamacppArgs
       }
     ]);
@@ -126,7 +152,7 @@ async function runSetupWizard(isEdit = false) {
     finalLlamacppArgs = '';
   }
   
-  // Q7: Custom llama.cpp build
+  // Q8: Custom llama.cpp build
   const existingCustomPath = existingConfig.customLlamacppPath || '';
   const hasExistingBuild = existingCustomPath.length > 0;
   
@@ -209,6 +235,7 @@ async function runSetupWizard(isEdit = false) {
     host,
     port: parseInt(port),
     logLevel,
+    contextSize,
     backend,
     modelDir: finalModelDir,
     runMode,

package/src/config/constants.js

@@ -57,6 +57,17 @@ const HOSTS = {
   ALL_INTERFACES: '0.0.0.0'
 };
 
+// Context window sizes (in tokens)
+const CONTEXT_SIZES = {
+  '4K': 4096,
+  '8K': 8192,
+  '16K': 16384,
+  '32K': 32768,
+  '64K': 65536,
+  '128K': 131072,
+  '256K': 262144
+};
+
 // Default values
 const DEFAULTS = {
   PORT: 8080,
@@ -65,7 +76,8 @@ const DEFAULTS = {
   EXPOSE_TO_NETWORK: false,
   RUN_MODE: RUN_MODES.SYSTEM_TRAY,
   MODEL_DIR: 'None',
-  BACKEND: BACKEND_TYPES.AUTO
+  BACKEND: BACKEND_TYPES.AUTO,
+  CONTEXT_SIZE: CONTEXT_SIZES['4K']
 };
 
 module.exports = {
@@ -79,5 +91,6 @@ module.exports = {
   LOG_LEVELS,
   RUN_MODES,
   HOSTS,
+  CONTEXT_SIZES,
   DEFAULTS
 };

package/src/services/server.js

@@ -1,11 +1,14 @@
 const fs = require('fs');
 const { execSync, spawn } = require('child_process');
+const kill = require('tree-kill');
 const { LEMONADE_SERVER_DEFAULT_PATH } = require('../config/constants');
 const { findLlamaServer } = require('../utils/system');
 const { getLlamaServerPath } = require('./asset-manager');
 
 // Track the server process for graceful shutdown
 let serverProcess = null;
+let isShuttingDown = false;
+let hasExited = false;
 
 /**
  * Build server command arguments
@@ -13,7 +16,7 @@ let serverProcess = null;
  * @returns {Array} Array of command arguments
  */
 function buildServerArgs(config) {
-  const { host, port, logLevel, modelDir, llamacppArgs } = config;
+  const { host, port, logLevel, modelDir, llamacppArgs, contextSize } = config;
   const args = [
     'serve',
     '--log-level', logLevel || 'info',
@@ -21,6 +24,10 @@ function buildServerArgs(config) {
     '--port', port.toString()
   ];
   
+  if (contextSize) {
+    args.push('--ctx-size', contextSize.toString());
+  }
+  
   if (modelDir && modelDir !== 'None') {
     args.push('--extra-models-dir', modelDir);
   }
@@ -76,6 +83,7 @@ async function launchLemonadeServer(config) {
     host, 
     port, 
     logLevel, 
+    contextSize,
     modelDir, 
     llamacppArgs,
     runMode,
@@ -89,6 +97,7 @@ async function launchLemonadeServer(config) {
   console.log(`Host: ${host}`);
   console.log(`Port: ${port}`);
   console.log(`Log Level: ${logLevel}`);
+  console.log(`Context Size: ${contextSize || 'default'}`);
   console.log(`Backend: ${backend || 'auto'}`);
   console.log(`Model Directory: ${modelDir || 'default'}`);
   console.log(`Run Mode: ${runMode || 'headless'}`);
@@ -175,8 +184,11 @@ async function launchLemonadeServer(config) {
     
     // Wait for the process to exit (blocking)
     await new Promise((resolve, reject) => {
-      serverProcess.on('close', (code) => {
-        console.log(`\nServer exited with code ${code}`);
+      serverProcess.on('exit', (code, signal) => {
+        // Only log if shutdown wasn't initiated by us
+        if (!hasExited) {
+          console.log(`\nServer exited with status ${code || 'None'} and signal ${signal || 'None'}`);
+        }
         serverProcess = null;
         resolve(code);
       });
@@ -197,37 +209,95 @@ async function launchLemonadeServer(config) {
 }
 
 /**
- * Gracefully shutdown the lemonade server
+ * Gracefully shutdown the lemonade server and all child processes
  */
 function shutdownLemonadeServer() {
-  if (serverProcess) {
-    console.log('\n\nShutting down Lemonade Server...');
-    
-    // Try graceful shutdown first (SIGTERM)
-    serverProcess.on('exit', () => {
-      console.log('Server shut down successfully.');
-    });
-    
-    serverProcess.on('error', (err) => {
-      console.error(`Error shutting down server: ${err.message}`);
-    });
-    
-    // Send SIGTERM signal
-    serverProcess.kill('SIGTERM');
-    
-    // Force kill after 5 seconds if still running
-    setTimeout(() => {
-      if (serverProcess && !serverProcess.killed) {
-        console.log('Force killing server...');
-        serverProcess.kill('SIGKILL');
-      }
-    }, 5000);
+  // Prevent duplicate shutdown calls
+  if (isShuttingDown) {
+    console.log('Shutdown already in progress...');
+    return;
+  }
+  
+  if (!serverProcess || !serverProcess.pid) {
+    console.log('No server process to shut down.');
+    return;
+  }
+  
+  isShuttingDown = true;
+  hasExited = false;
+  console.log('\n\nShutting down Lemonade Server and child processes...');
+  
+  // Check if process is still running
+  try {
+    // Sending signal 0 checks if process exists without actually sending a signal
+    process.kill(serverProcess.pid, 0);
+  } catch (err) {
+    // Process is already dead
+    console.log('Server process is already terminated.');
+    serverProcess = null;
+    isShuttingDown = false;
+    return;
   }
+  
+  // Remove existing event listeners to prevent duplicates
+  serverProcess.removeAllListeners('exit');
+  serverProcess.removeAllListeners('error');
+  
+  // Try graceful shutdown first (SIGINT)
+  serverProcess.on('exit', (code, signal) => {
+    hasExited = true;
+    console.log(`Server exited with status ${code || 'None'} and signal ${signal || 'None'}`);
+    console.log('Server shut down successfully.');
+    serverProcess = null;
+    isShuttingDown = false;
+  });
+  
+  serverProcess.on('error', (err) => {
+    console.error(`Error shutting down server: ${err.message}`);
+  });
+  
+  // Use tree-kill to terminate the process and all its children
+  kill(serverProcess.pid, 'SIGINT', (err) => {
+    if (err && !hasExited) {
+      // Only log error if process hasn't already exited naturally
+      if (err.code !== 'ESRCH') {
+        console.log(`Note: Could not kill process tree: ${err.message}`);
+      }
+    }
+  });
+  
+  // Force kill with SIGKILL after 3 seconds if still running
+  setTimeout(() => {
+    // Don't force kill if already exited
+    if (hasExited || !isShuttingDown) return;
+    
+    // Check again if process is still running before force kill
+    try {
+      process.kill(serverProcess.pid, 0);
+      // Process still exists, force kill it
+      kill(serverProcess.pid, 'SIGKILL', (err) => {
+        if (err && !hasExited) {
+          // Only log error if process hasn't already exited naturally
+          if (err.code !== 'ESRCH') {
+            console.log(`Note: Could not force kill process: ${err.message}`);
+          }
+        }
+      });
+    } catch (err) {
+      // Process has already exited
+      console.log('Process has already exited.');
+    }
+  }, 3000);
 }
 
 // Set up signal handlers for graceful shutdown
 function setupShutdownHandlers() {
   const shutdown = (signal) => {
+    if (isShuttingDown) {
+      console.log(`Received ${signal}, but shutdown already in progress...`);
+      return;
+    }
+    
     console.log(`\n\nReceived ${signal}. Shutting down...`);
     shutdownLemonadeServer();
     setTimeout(() => process.exit(0), 2000);