Merge pull request #1888 from janhq/dev

Release cut 0.4.6

.devcontainer/devcontainer.json

@@ -0,0 +1,4 @@
+{
+    "name": "jan",
+    "image": "node:20"
+}

.github/workflows/jan-electron-build-nightly.yml

@@ -8,7 +8,7 @@ on:
       - 'README.md'
       - 'docs/**'
   schedule:
-    - cron: '0 20 * * 2,3,4' # At 8 PM UTC on Tuesday, Wednesday, and Thursday, which is 3 AM UTC+7
+    - cron: '0 20 * * 1,2,3' # At 8 PM UTC on Monday, Tuesday, and Wednesday which is 3 AM UTC+7 Tuesday, Wednesday, and Thursday
   workflow_dispatch:
     inputs:
       public_provider:

.github/workflows/template-build-linux-x64.yml

@@ -98,8 +98,8 @@ jobs:
           make build-and-publish
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          ANALYTICS_ID: ${{ secrets.JAN_APP_POSTHOG_PROJECT_API_KEY }}
+          ANALYTICS_ID: ${{ secrets.JAN_APP_UMAMI_PROJECT_API_KEY }}
-          ANALYTICS_HOST: ${{ secrets.JAN_APP_POSTHOG_URL }}
+          ANALYTICS_HOST: ${{ secrets.JAN_APP_UMAMI_URL }}
 
       - name: Upload Artifact .deb file
         if: inputs.public_provider != 'github'

.github/workflows/template-build-macos.yml

@@ -137,8 +137,8 @@ jobs:
           APPLE_APP_SPECIFIC_PASSWORD: ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }}
           APP_PATH: "."
           DEVELOPER_ID: ${{ secrets.DEVELOPER_ID }}
-          ANALYTICS_ID: ${{ secrets.JAN_APP_POSTHOG_PROJECT_API_KEY }}
+          ANALYTICS_ID: ${{ secrets.JAN_APP_UMAMI_PROJECT_API_KEY }}
-          ANALYTICS_HOST: ${{ secrets.JAN_APP_POSTHOG_URL }}
+          ANALYTICS_HOST: ${{ secrets.JAN_APP_UMAMI_URL }}
 
       - name: Upload Artifact
         if: inputs.public_provider != 'github'

.github/workflows/template-build-windows-x64.yml

@@ -127,8 +127,8 @@ jobs:
           make build-and-publish
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          ANALYTICS_ID: ${{ secrets.JAN_APP_POSTHOG_PROJECT_API_KEY }}
+          ANALYTICS_ID: ${{ secrets.JAN_APP_UMAMI_PROJECT_API_KEY }}
-          ANALYTICS_HOST: ${{ secrets.JAN_APP_POSTHOG_URL }}
+          ANALYTICS_HOST: ${{ secrets.JAN_APP_UMAMI_URL }}
           AZURE_KEY_VAULT_URI: ${{ secrets.AZURE_KEY_VAULT_URI }}
           AZURE_CLIENT_ID: ${{ secrets.AZURE_CLIENT_ID }}
           AZURE_TENANT_ID: ${{ secrets.AZURE_TENANT_ID }}

.github/workflows/update-release-url.yml

@@ -17,7 +17,7 @@ jobs:
         with:
           fetch-depth: "0"
           token: ${{ secrets.PAT_SERVICE_ACCOUNT }}
-          ref: main
+          ref: dev
 
       - name: Get Latest Release
         uses: pozetroninc/github-action-get-latest-release@v0.7.0
@@ -46,4 +46,4 @@ jobs:
           git config --global user.name "Service Account"
           git add README.md
           git commit -m "Update README.md with Stable Download URLs"
-          git -c http.extraheader="AUTHORIZATION: bearer ${{ secrets.PAT_SERVICE_ACCOUNT }}" push origin HEAD:main
+          git -c http.extraheader="AUTHORIZATION: bearer ${{ secrets.PAT_SERVICE_ACCOUNT }}" push origin HEAD:dev

.gitignore

@@ -12,6 +12,8 @@ build
 electron/renderer
 electron/models
 electron/docs
+electron/engines
+server/pre-install
 package-lock.json
 
 *.log
@@ -26,3 +28,4 @@ extensions/inference-nitro-extension/bin/*/*.exp
 extensions/inference-nitro-extension/bin/*/*.lib
 extensions/inference-nitro-extension/bin/saved-*
 extensions/inference-nitro-extension/bin/*.tar.gz
+

README.md

@@ -76,31 +76,31 @@ Jan is an open-source ChatGPT alternative that runs 100% offline on your compute
   <tr style="text-align:center">
     <td style="text-align:center"><b>Experimental (Nightly Build)</b></td>
     <td style="text-align:center">
-      <a href='https://delta.jan.ai/latest/jan-win-x64-0.4.4-170.exe'>
+      <a href='https://delta.jan.ai/latest/jan-win-x64-0.4.5-216.exe'>
         <img src='./docs/static/img/windows.png' style="height:14px; width: 14px" />
         <b>jan.exe</b>
       </a>
     </td>
     <td style="text-align:center">
-      <a href='https://delta.jan.ai/latest/jan-mac-x64-0.4.4-170.dmg'>
+      <a href='https://delta.jan.ai/latest/jan-mac-x64-0.4.5-216.dmg'>
         <img src='./docs/static/img/mac.png' style="height:15px; width: 15px" />
         <b>Intel</b>
       </a>
     </td>
     <td style="text-align:center">
-      <a href='https://delta.jan.ai/latest/jan-mac-arm64-0.4.4-170.dmg'>
+      <a href='https://delta.jan.ai/latest/jan-mac-arm64-0.4.5-216.dmg'>
         <img src='./docs/static/img/mac.png' style="height:15px; width: 15px" />
         <b>M1/M2</b>
       </a>
     </td>
     <td style="text-align:center">
-      <a href='https://delta.jan.ai/latest/jan-linux-amd64-0.4.4-170.deb'>
+      <a href='https://delta.jan.ai/latest/jan-linux-amd64-0.4.5-216.deb'>
         <img src='./docs/static/img/linux.png' style="height:14px; width: 14px" />
         <b>jan.deb</b>
       </a>
     </td>
     <td style="text-align:center">
-      <a href='https://delta.jan.ai/latest/jan-linux-x86_64-0.4.4-170.AppImage'>
+      <a href='https://delta.jan.ai/latest/jan-linux-x86_64-0.4.5-216.AppImage'>
         <img src='./docs/static/img/linux.png' style="height:14px; width: 14px" />
         <b>jan.AppImage</b>
       </a>

core/.prettierignore

@@ -0,0 +1,5 @@
+.next/
+node_modules/
+dist/
+*.hbs
+*.mdx

core/src/api/index.ts

@@ -3,7 +3,6 @@
  * @description Enum of all the routes exposed by the app
  */
 export enum AppRoute {
-  appDataPath = 'appDataPath',
   openExternalUrl = 'openExternalUrl',
   openAppDirectory = 'openAppDirectory',
   openFileExplore = 'openFileExplorer',
@@ -12,6 +11,7 @@ export enum AppRoute {
   updateAppConfiguration = 'updateAppConfiguration',
   relaunch = 'relaunch',
   joinPath = 'joinPath',
+  isSubdirectory = 'isSubdirectory',
   baseName = 'baseName',
   startServer = 'startServer',
   stopServer = 'stopServer',
@@ -61,7 +61,9 @@ export enum FileManagerRoute {
   syncFile = 'syncFile',
   getJanDataFolderPath = 'getJanDataFolderPath',
   getResourcePath = 'getResourcePath',
+  getUserHomePath = 'getUserHomePath',
   fileStat = 'fileStat',
+  writeBlob = 'writeBlob',
 }
 
 export type ApiFunction = (...args: any[]) => any

core/src/core.ts

@@ -22,7 +22,11 @@ const executeOnMain: (extension: string, method: string, ...args: any[]) => Prom
  * @param {object} network - Optional object to specify proxy/whether to ignore SSL certificates.
  * @returns {Promise<any>} A promise that resolves when the file is downloaded.
  */
-const downloadFile: (url: string, fileName: string, network?: { proxy?: string, ignoreSSL?: boolean }) => Promise<any> = (url, fileName, network) => {
+const downloadFile: (
+  url: string,
+  fileName: string,
+  network?: { proxy?: string; ignoreSSL?: boolean }
+) => Promise<any> = (url, fileName, network) => {
   return global.core?.api?.downloadFile(url, fileName, network)
 }
 
@@ -79,6 +83,12 @@ const openExternalUrl: (url: string) => Promise<any> = (url) =>
  */
 const getResourcePath: () => Promise<string> = () => global.core.api?.getResourcePath()
 
+/**
+ * Gets the user's home path.
+ * @returns return user's home path
+ */
+const getUserHomePath = (): Promise<string> => global.core.api?.getUserHomePath()
+
 /**
  * Log to file from browser processes.
  *
@@ -87,6 +97,17 @@ const getResourcePath: () => Promise<string> = () => global.core.api?.getResourc
 const log: (message: string, fileName?: string) => void = (message, fileName) =>
   global.core.api?.log(message, fileName)
 
+/**
+ * Check whether the path is a subdirectory of another path.
+ *
+ * @param from - The path to check.
+ * @param to - The path to check against.
+ *
+ * @returns {Promise<boolean>} - A promise that resolves with a boolean indicating whether the path is a subdirectory.
+ */
+const isSubdirectory: (from: string, to: string) => Promise<boolean> = (from: string, to: string) =>
+  global.core.api?.isSubdirectory(from, to)
+
 /**
  * Register extension point function type definition
  */
@@ -94,7 +115,7 @@ export type RegisterExtensionPoint = (
   extensionName: string,
   extensionId: string,
   method: Function,
-  priority?: number,
+  priority?: number
 ) => void
 
 /**
@@ -111,5 +132,7 @@ export {
   openExternalUrl,
   baseName,
   log,
+  isSubdirectory,
+  getUserHomePath,
   FileStat,
 }

core/src/fs.ts

@@ -1,4 +1,4 @@
-import { FileStat } from "./types"
+import { FileStat } from './types'
 
 /**
  * Writes data to a file at the specified path.
@@ -6,6 +6,15 @@ import { FileStat } from "./types"
  */
 const writeFileSync = (...args: any[]) => global.core.api?.writeFileSync(...args)
 
+/**
+ * Writes blob data to a file at the specified path.
+ * @param path - The path to file.
+ * @param data - The blob data.
+ * @returns
+ */
+const writeBlob: (path: string, data: string) => Promise<any> = (path, data) =>
+  global.core.api?.writeBlob(path, data)
+
 /**
  * Reads the contents of a file at the specified path.
  * @returns {Promise<any>} A Promise that resolves with the contents of the file.
@@ -60,7 +69,6 @@ const syncFile: (src: string, dest: string) => Promise<any> = (src, dest) =>
  */
 const copyFileSync = (...args: any[]) => global.core.api?.copyFileSync(...args)
 
-
 /**
  * Gets the file's stats.
  *
@@ -70,7 +78,6 @@ const copyFileSync = (...args: any[]) => global.core.api?.copyFileSync(...args)
 const fileStat: (path: string) => Promise<FileStat | undefined> = (path) =>
   global.core.api?.fileStat(path)
 
-
 // TODO: Export `dummy` fs functions automatically
 // Currently adding these manually
 export const fs = {
@@ -84,5 +91,6 @@ export const fs = {
   appendFileSync,
   copyFileSync,
   syncFile,
-  fileStat
+  fileStat,
+  writeBlob,
 }

core/src/node/api/common/builder.ts

@@ -2,7 +2,8 @@ import fs from 'fs'
 import { JanApiRouteConfiguration, RouteConfiguration } from './configuration'
 import { join } from 'path'
 import { ContentType, MessageStatus, Model, ThreadMessage } from './../../../index'
-import { getJanDataFolderPath } from '../../utils'
+import { getEngineConfiguration, getJanDataFolderPath } from '../../utils'
+import { DEFAULT_CHAT_COMPLETION_URL } from './consts'
 
 export const getBuilder = async (configuration: RouteConfiguration) => {
   const directoryPath = join(getJanDataFolderPath(), configuration.dirName)
@@ -265,19 +266,22 @@ export const downloadModel = async (
   const modelBinaryPath = join(directoryPath, modelId)
 
   const request = require('request')
-  const rq = request({ url: model.source_url, strictSSL, proxy })
   const progress = require('request-progress')
+
+  for (const source of model.sources) {
+    const rq = request({ url: source, strictSSL, proxy })
     progress(rq, {})
       .on('progress', function (state: any) {
-      console.log('progress', JSON.stringify(state, null, 2))
+        console.debug('progress', JSON.stringify(state, null, 2))
       })
       .on('error', function (err: Error) {
         console.error('error', err)
       })
       .on('end', function () {
-      console.log('end')
+        console.debug('end')
       })
       .pipe(fs.createWriteStream(modelBinaryPath))
+  }
 
   return {
     message: `Starting download ${modelId}`,
@@ -306,7 +310,7 @@ export const chatCompletions = async (request: any, reply: any) => {
   const engineConfiguration = await getEngineConfiguration(requestedModel.engine)
 
   let apiKey: string | undefined = undefined
-  let apiUrl: string = 'http://127.0.0.1:3928/inferences/llamacpp/chat_completion' // default nitro url
+  let apiUrl: string = DEFAULT_CHAT_COMPLETION_URL
 
   if (engineConfiguration) {
     apiKey = engineConfiguration.api_key
@@ -317,7 +321,7 @@ export const chatCompletions = async (request: any, reply: any) => {
     'Content-Type': 'text/event-stream',
     'Cache-Control': 'no-cache',
     'Connection': 'keep-alive',
-    "Access-Control-Allow-Origin": "*"
+    'Access-Control-Allow-Origin': '*',
   })
 
   const headers: Record<string, any> = {
@@ -343,13 +347,3 @@ export const chatCompletions = async (request: any, reply: any) => {
     response.body.pipe(reply.raw)
   }
 }
-
-const getEngineConfiguration = async (engineId: string) => {
-  if (engineId !== 'openai') {
-    return undefined
-  }
-  const directoryPath = join(getJanDataFolderPath(), 'engines')
-  const filePath = join(directoryPath, `${engineId}.json`)
-  const data = await fs.readFileSync(filePath, 'utf-8')
-  return JSON.parse(data)
-}

core/src/node/api/common/consts.ts

@@ -0,0 +1,19 @@
+// The PORT to use for the Nitro subprocess
+export const NITRO_DEFAULT_PORT = 3928
+
+// The HOST address to use for the Nitro subprocess
+export const LOCAL_HOST = '127.0.0.1'
+
+export const SUPPORTED_MODEL_FORMAT = '.gguf'
+
+// The URL for the Nitro subprocess
+const NITRO_HTTP_SERVER_URL = `http://${LOCAL_HOST}:${NITRO_DEFAULT_PORT}`
+// The URL for the Nitro subprocess to load a model
+export const NITRO_HTTP_LOAD_MODEL_URL = `${NITRO_HTTP_SERVER_URL}/inferences/llamacpp/loadmodel`
+// The URL for the Nitro subprocess to validate a model
+export const NITRO_HTTP_VALIDATE_MODEL_URL = `${NITRO_HTTP_SERVER_URL}/inferences/llamacpp/modelstatus`
+
+// The URL for the Nitro subprocess to kill itself
+export const NITRO_HTTP_KILL_URL = `${NITRO_HTTP_SERVER_URL}/processmanager/destroy`
+
+export const DEFAULT_CHAT_COMPLETION_URL = `http://${LOCAL_HOST}:${NITRO_DEFAULT_PORT}/inferences/llamacpp/chat_completion` // default nitro url

core/src/node/api/common/startStopModel.ts

@@ -0,0 +1,351 @@
+import fs from 'fs'
+import { join } from 'path'
+import { getJanDataFolderPath, getJanExtensionsPath, getSystemResourceInfo } from '../../utils'
+import { logServer } from '../../log'
+import { ChildProcessWithoutNullStreams, spawn } from 'child_process'
+import { Model, ModelSettingParams, PromptTemplate } from '../../../types'
+import {
+  LOCAL_HOST,
+  NITRO_DEFAULT_PORT,
+  NITRO_HTTP_KILL_URL,
+  NITRO_HTTP_LOAD_MODEL_URL,
+  NITRO_HTTP_VALIDATE_MODEL_URL,
+  SUPPORTED_MODEL_FORMAT,
+} from './consts'
+
+// The subprocess instance for Nitro
+let subprocess: ChildProcessWithoutNullStreams | undefined = undefined
+
+// TODO: move this to core type
+interface NitroModelSettings extends ModelSettingParams {
+  llama_model_path: string
+  cpu_threads: number
+}
+
+export const startModel = async (modelId: string, settingParams?: ModelSettingParams) => {
+  try {
+    await runModel(modelId, settingParams)
+
+    return {
+      message: `Model ${modelId} started`,
+    }
+  } catch (e) {
+    return {
+      error: e,
+    }
+  }
+}
+
+const runModel = async (modelId: string, settingParams?: ModelSettingParams): Promise<void> => {
+  const janDataFolderPath = getJanDataFolderPath()
+  const modelFolderFullPath = join(janDataFolderPath, 'models', modelId)
+
+  if (!fs.existsSync(modelFolderFullPath)) {
+    throw `Model not found: ${modelId}`
+  }
+
+  const files: string[] = fs.readdirSync(modelFolderFullPath)
+
+  // Look for GGUF model file
+  const ggufBinFile = files.find((file) => file.toLowerCase().includes(SUPPORTED_MODEL_FORMAT))
+
+  const modelMetadataPath = join(modelFolderFullPath, 'model.json')
+  const modelMetadata: Model = JSON.parse(fs.readFileSync(modelMetadataPath, 'utf-8'))
+
+  if (!ggufBinFile) {
+    throw 'No GGUF model file found'
+  }
+  const modelBinaryPath = join(modelFolderFullPath, ggufBinFile)
+
+  const nitroResourceProbe = await getSystemResourceInfo()
+  const nitroModelSettings: NitroModelSettings = {
+    ...modelMetadata.settings,
+    ...settingParams,
+    llama_model_path: modelBinaryPath,
+    // This is critical and requires real CPU physical core count (or performance core)
+    cpu_threads: Math.max(1, nitroResourceProbe.numCpuPhysicalCore),
+    ...(modelMetadata.settings.mmproj && {
+      mmproj: join(modelFolderFullPath, modelMetadata.settings.mmproj),
+    }),
+  }
+
+  logServer(`[NITRO]::Debug: Nitro model settings: ${JSON.stringify(nitroModelSettings)}`)
+
+  // Convert settings.prompt_template to system_prompt, user_prompt, ai_prompt
+  if (modelMetadata.settings.prompt_template) {
+    const promptTemplate = modelMetadata.settings.prompt_template
+    const prompt = promptTemplateConverter(promptTemplate)
+    if (prompt?.error) {
+      return Promise.reject(prompt.error)
+    }
+    nitroModelSettings.system_prompt = prompt.system_prompt
+    nitroModelSettings.user_prompt = prompt.user_prompt
+    nitroModelSettings.ai_prompt = prompt.ai_prompt
+  }
+
+  await runNitroAndLoadModel(modelId, nitroModelSettings)
+}
+
+// TODO: move to util
+const promptTemplateConverter = (promptTemplate: string): PromptTemplate => {
+  // Split the string using the markers
+  const systemMarker = '{system_message}'
+  const promptMarker = '{prompt}'
+
+  if (promptTemplate.includes(systemMarker) && promptTemplate.includes(promptMarker)) {
+    // Find the indices of the markers
+    const systemIndex = promptTemplate.indexOf(systemMarker)
+    const promptIndex = promptTemplate.indexOf(promptMarker)
+
+    // Extract the parts of the string
+    const system_prompt = promptTemplate.substring(0, systemIndex)
+    const user_prompt = promptTemplate.substring(systemIndex + systemMarker.length, promptIndex)
+    const ai_prompt = promptTemplate.substring(promptIndex + promptMarker.length)
+
+    // Return the split parts
+    return { system_prompt, user_prompt, ai_prompt }
+  } else if (promptTemplate.includes(promptMarker)) {
+    // Extract the parts of the string for the case where only promptMarker is present
+    const promptIndex = promptTemplate.indexOf(promptMarker)
+    const user_prompt = promptTemplate.substring(0, promptIndex)
+    const ai_prompt = promptTemplate.substring(promptIndex + promptMarker.length)
+
+    // Return the split parts
+    return { user_prompt, ai_prompt }
+  }
+
+  // Return an error if none of the conditions are met
+  return { error: 'Cannot split prompt template' }
+}
+
+const runNitroAndLoadModel = async (modelId: string, modelSettings: NitroModelSettings) => {
+  // Gather system information for CPU physical cores and memory
+  const tcpPortUsed = require('tcp-port-used')
+
+  await stopModel(modelId)
+  await tcpPortUsed.waitUntilFree(NITRO_DEFAULT_PORT, 300, 5000)
+
+  /**
+   * There is a problem with Windows process manager
+   * Should wait for awhile to make sure the port is free and subprocess is killed
+   * The tested threshold is 500ms
+   **/
+  if (process.platform === 'win32') {
+    await new Promise((resolve) => setTimeout(resolve, 500))
+  }
+
+  await spawnNitroProcess()
+  await loadLLMModel(modelSettings)
+  await validateModelStatus()
+}
+
+const spawnNitroProcess = async (): Promise<void> => {
+  logServer(`[NITRO]::Debug: Spawning Nitro subprocess...`)
+
+  let binaryFolder = join(
+    getJanExtensionsPath(),
+    '@janhq',
+    'inference-nitro-extension',
+    'dist',
+    'bin'
+  )
+
+  let executableOptions = executableNitroFile()
+  const tcpPortUsed = require('tcp-port-used')
+
+  const args: string[] = ['1', LOCAL_HOST, NITRO_DEFAULT_PORT.toString()]
+  // Execute the binary
+  logServer(
+    `[NITRO]::Debug: Spawn nitro at path: ${executableOptions.executablePath}, and args: ${args}`
+  )
+  subprocess = spawn(
+    executableOptions.executablePath,
+    ['1', LOCAL_HOST, NITRO_DEFAULT_PORT.toString()],
+    {
+      cwd: binaryFolder,
+      env: {
+        ...process.env,
+        CUDA_VISIBLE_DEVICES: executableOptions.cudaVisibleDevices,
+      },
+    }
+  )
+
+  // Handle subprocess output
+  subprocess.stdout.on('data', (data: any) => {
+    logServer(`[NITRO]::Debug: ${data}`)
+  })
+
+  subprocess.stderr.on('data', (data: any) => {
+    logServer(`[NITRO]::Error: ${data}`)
+  })
+
+  subprocess.on('close', (code: any) => {
+    logServer(`[NITRO]::Debug: Nitro exited with code: ${code}`)
+    subprocess = undefined
+  })
+
+  tcpPortUsed.waitUntilUsed(NITRO_DEFAULT_PORT, 300, 30000).then(() => {
+    logServer(`[NITRO]::Debug: Nitro is ready`)
+  })
+}
+
+type NitroExecutableOptions = {
+  executablePath: string
+  cudaVisibleDevices: string
+}
+
+const executableNitroFile = (): NitroExecutableOptions => {
+  const nvidiaInfoFilePath = join(getJanDataFolderPath(), 'settings', 'settings.json')
+  let binaryFolder = join(
+    getJanExtensionsPath(),
+    '@janhq',
+    'inference-nitro-extension',
+    'dist',
+    'bin'
+  )
+
+  let cudaVisibleDevices = ''
+  let binaryName = 'nitro'
+  /**
+   * The binary folder is different for each platform.
+   */
+  if (process.platform === 'win32') {
+    /**
+     *  For Windows: win-cpu, win-cuda-11-7, win-cuda-12-0
+     */
+    let nvidiaInfo = JSON.parse(fs.readFileSync(nvidiaInfoFilePath, 'utf-8'))
+    if (nvidiaInfo['run_mode'] === 'cpu') {
+      binaryFolder = join(binaryFolder, 'win-cpu')
+    } else {
+      if (nvidiaInfo['cuda'].version === '12') {
+        binaryFolder = join(binaryFolder, 'win-cuda-12-0')
+      } else {
+        binaryFolder = join(binaryFolder, 'win-cuda-11-7')
+      }
+      cudaVisibleDevices = nvidiaInfo['gpu_highest_vram']
+    }
+    binaryName = 'nitro.exe'
+  } else if (process.platform === 'darwin') {
+    /**
+     *  For MacOS: mac-arm64 (Silicon), mac-x64 (InteL)
+     */
+    if (process.arch === 'arm64') {
+      binaryFolder = join(binaryFolder, 'mac-arm64')
+    } else {
+      binaryFolder = join(binaryFolder, 'mac-x64')
+    }
+  } else {
+    /**
+     *  For Linux: linux-cpu, linux-cuda-11-7, linux-cuda-12-0
+     */
+    let nvidiaInfo = JSON.parse(fs.readFileSync(nvidiaInfoFilePath, 'utf-8'))
+    if (nvidiaInfo['run_mode'] === 'cpu') {
+      binaryFolder = join(binaryFolder, 'linux-cpu')
+    } else {
+      if (nvidiaInfo['cuda'].version === '12') {
+        binaryFolder = join(binaryFolder, 'linux-cuda-12-0')
+      } else {
+        binaryFolder = join(binaryFolder, 'linux-cuda-11-7')
+      }
+      cudaVisibleDevices = nvidiaInfo['gpu_highest_vram']
+    }
+  }
+
+  return {
+    executablePath: join(binaryFolder, binaryName),
+    cudaVisibleDevices,
+  }
+}
+
+const validateModelStatus = async (): Promise<void> => {
+  // Send a GET request to the validation URL.
+  // Retry the request up to 3 times if it fails, with a delay of 500 milliseconds between retries.
+  const fetchRT = require('fetch-retry')
+  const fetchRetry = fetchRT(fetch)
+
+  return fetchRetry(NITRO_HTTP_VALIDATE_MODEL_URL, {
+    method: 'GET',
+    headers: {
+      'Content-Type': 'application/json',
+    },
+    retries: 5,
+    retryDelay: 500,
+  }).then(async (res: Response) => {
+    logServer(`[NITRO]::Debug: Validate model state success with response ${JSON.stringify(res)}`)
+    // If the response is OK, check model_loaded status.
+    if (res.ok) {
+      const body = await res.json()
+      // If the model is loaded, return an empty object.
+      // Otherwise, return an object with an error message.
+      if (body.model_loaded) {
+        return Promise.resolve()
+      }
+    }
+    return Promise.reject('Validate model status failed')
+  })
+}
+
+const loadLLMModel = async (settings: NitroModelSettings): Promise<Response> => {
+  logServer(`[NITRO]::Debug: Loading model with params ${JSON.stringify(settings)}`)
+  const fetchRT = require('fetch-retry')
+  const fetchRetry = fetchRT(fetch)
+
+  return fetchRetry(NITRO_HTTP_LOAD_MODEL_URL, {
+    method: 'POST',
+    headers: {
+      'Content-Type': 'application/json',
+    },
+    body: JSON.stringify(settings),
+    retries: 3,
+    retryDelay: 500,
+  })
+    .then((res: any) => {
+      logServer(`[NITRO]::Debug: Load model success with response ${JSON.stringify(res)}`)
+      return Promise.resolve(res)
+    })
+    .catch((err: any) => {
+      logServer(`[NITRO]::Error: Load model failed with error ${err}`)
+      return Promise.reject(err)
+    })
+}
+
+/**
+ * Stop model and kill nitro process.
+ */
+export const stopModel = async (_modelId: string) => {
+  if (!subprocess) {
+    return {
+      error: "Model isn't running",
+    }
+  }
+  return new Promise((resolve, reject) => {
+    const controller = new AbortController()
+    setTimeout(() => {
+      controller.abort()
+      reject({
+        error: 'Failed to stop model: Timedout',
+      })
+    }, 5000)
+    const tcpPortUsed = require('tcp-port-used')
+    logServer(`[NITRO]::Debug: Request to kill Nitro`)
+
+    fetch(NITRO_HTTP_KILL_URL, {
+      method: 'DELETE',
+      signal: controller.signal,
+    })
+      .then(() => {
+        subprocess?.kill()
+        subprocess = undefined
+      })
+      .catch(() => {
+        // don't need to do anything, we still kill the subprocess
+      })
+      .then(() => tcpPortUsed.waitUntilFree(NITRO_DEFAULT_PORT, 300, 5000))
+      .then(() => logServer(`[NITRO]::Debug: Nitro process is terminated`))
+      .then(() =>
+        resolve({
+          message: 'Model stopped',
+        })
+      )
+  })
+}

core/src/node/api/routes/common.ts

@@ -10,6 +10,8 @@ import {
 } from '../common/builder'
 
 import { JanApiRouteConfiguration } from '../common/configuration'
+import { startModel, stopModel } from '../common/startStopModel'
+import { ModelSettingParams } from '../../../types'
 
 export const commonRouter = async (app: HttpServer) => {
   // Common Routes
@@ -17,19 +19,33 @@ export const commonRouter = async (app: HttpServer) => {
     app.get(`/${key}`, async (_request) => getBuilder(JanApiRouteConfiguration[key]))
 
     app.get(`/${key}/:id`, async (request: any) =>
-      retrieveBuilder(JanApiRouteConfiguration[key], request.params.id),
+      retrieveBuilder(JanApiRouteConfiguration[key], request.params.id)
     )
 
     app.delete(`/${key}/:id`, async (request: any) =>
-      deleteBuilder(JanApiRouteConfiguration[key], request.params.id),
+      deleteBuilder(JanApiRouteConfiguration[key], request.params.id)
     )
   })
 
   // Download Model Routes
   app.get(`/models/download/:modelId`, async (request: any) =>
-    downloadModel(request.params.modelId, { ignoreSSL: request.query.ignoreSSL === 'true', proxy: request.query.proxy }),
+    downloadModel(request.params.modelId, {
+      ignoreSSL: request.query.ignoreSSL === 'true',
+      proxy: request.query.proxy,
+    })
   )
 
+  app.put(`/models/:modelId/start`, async (request: any) => {
+    let settingParams: ModelSettingParams | undefined = undefined
+    if (Object.keys(request.body).length !== 0) {
+      settingParams = JSON.parse(request.body) as ModelSettingParams
+    }
+
+    return startModel(request.params.modelId, settingParams)
+  })
+
+  app.put(`/models/:modelId/stop`, async (request: any) => stopModel(request.params.modelId))
+
   // Chat Completion Routes
   app.post(`/chat/completions`, async (request: any, reply: any) => chatCompletions(request, reply))
 

core/src/node/api/routes/fileManager.ts

@@ -8,5 +8,7 @@ export const fsRouter = async (app: HttpServer) => {
 
   app.post(`/app/${FileManagerRoute.getResourcePath}`, async (request: any, reply: any) => {})
 
+  app.post(`/app/${FileManagerRoute.getUserHomePath}`, async (request: any, reply: any) => {})
+
   app.post(`/app/${FileManagerRoute.fileStat}`, async (request: any, reply: any) => {})
 }

core/src/node/api/routes/fs.ts

@@ -2,6 +2,7 @@ import { FileSystemRoute } from '../../../api'
 import { join } from 'path'
 import { HttpServer } from '../HttpServer'
 import { getJanDataFolderPath } from '../../utils'
+import { normalizeFilePath } from '../../path'
 
 export const fsRouter = async (app: HttpServer) => {
   const moduleName = 'fs'
@@ -13,10 +14,10 @@ export const fsRouter = async (app: HttpServer) => {
         const result = await import(moduleName).then((mdl) => {
           return mdl[route](
             ...body.map((arg: any) =>
-              typeof arg === 'string' && arg.includes('file:/')
+              typeof arg === 'string' && (arg.startsWith(`file:/`) || arg.startsWith(`file:\\`))
-                ? join(getJanDataFolderPath(), arg.replace('file:/', ''))
+                ? join(getJanDataFolderPath(), normalizeFilePath(arg))
-                : arg,
+                : arg
-            ),
+            )
           )
         })
         res.status(200).send(result)

core/src/node/utils/index.ts

@@ -1,16 +1,18 @@
-import { AppConfiguration } from "../../types";
+import { AppConfiguration, SystemResourceInfo } from '../../types'
-import { join } from "path";
+import { join } from 'path'
-import fs from "fs";
+import fs from 'fs'
-import os from "os";
+import os from 'os'
+import { log, logServer } from '../log'
+import childProcess from 'child_process'
 
 // TODO: move this to core
-const configurationFileName = "settings.json";
+const configurationFileName = 'settings.json'
 
 // TODO: do no specify app name in framework module
-const defaultJanDataFolder = join(os.homedir(), "jan");
+const defaultJanDataFolder = join(os.homedir(), 'jan')
 const defaultAppConfig: AppConfiguration = {
   data_folder: defaultJanDataFolder,
-};
+}
 
 /**
  * Getting App Configurations.
@@ -20,39 +22,39 @@ const defaultAppConfig: AppConfiguration = {
 export const getAppConfigurations = (): AppConfiguration => {
   // Retrieve Application Support folder path
   // Fallback to user home directory if not found
-  const configurationFile = getConfigurationFilePath();
+  const configurationFile = getConfigurationFilePath()
 
   if (!fs.existsSync(configurationFile)) {
     // create default app config if we don't have one
-    console.debug(`App config not found, creating default config at ${configurationFile}`);
+    console.debug(`App config not found, creating default config at ${configurationFile}`)
-    fs.writeFileSync(configurationFile, JSON.stringify(defaultAppConfig));
+    fs.writeFileSync(configurationFile, JSON.stringify(defaultAppConfig))
-    return defaultAppConfig;
+    return defaultAppConfig
   }
 
   try {
     const appConfigurations: AppConfiguration = JSON.parse(
-      fs.readFileSync(configurationFile, "utf-8"),
+      fs.readFileSync(configurationFile, 'utf-8')
-    );
+    )
-    return appConfigurations;
+    return appConfigurations
   } catch (err) {
-    console.error(`Failed to read app config, return default config instead! Err: ${err}`);
+    console.error(`Failed to read app config, return default config instead! Err: ${err}`)
-    return defaultAppConfig;
+    return defaultAppConfig
+  }
 }
-};
 
 const getConfigurationFilePath = () =>
   join(
-    global.core?.appPath() || process.env[process.platform == "win32" ? "USERPROFILE" : "HOME"],
+    global.core?.appPath() || process.env[process.platform == 'win32' ? 'USERPROFILE' : 'HOME'],
-    configurationFileName,
+    configurationFileName
-  );
+  )
 
 export const updateAppConfiguration = (configuration: AppConfiguration): Promise<void> => {
-  const configurationFile = getConfigurationFilePath();
+  const configurationFile = getConfigurationFilePath()
-  console.debug("updateAppConfiguration, configurationFile: ", configurationFile);
+  console.debug('updateAppConfiguration, configurationFile: ', configurationFile)
 
-  fs.writeFileSync(configurationFile, JSON.stringify(configuration));
+  fs.writeFileSync(configurationFile, JSON.stringify(configuration))
-  return Promise.resolve();
+  return Promise.resolve()
-};
+}
 
 /**
  * Utility function to get server log path
@@ -60,13 +62,13 @@ export const updateAppConfiguration = (configuration: AppConfiguration): Promise
  * @returns {string} The log path.
  */
 export const getServerLogPath = (): string => {
-  const appConfigurations = getAppConfigurations();
+  const appConfigurations = getAppConfigurations()
-  const logFolderPath = join(appConfigurations.data_folder, "logs");
+  const logFolderPath = join(appConfigurations.data_folder, 'logs')
   if (!fs.existsSync(logFolderPath)) {
-    fs.mkdirSync(logFolderPath, { recursive: true });
+    fs.mkdirSync(logFolderPath, { recursive: true })
+  }
+  return join(logFolderPath, 'server.log')
 }
-  return join(logFolderPath, "server.log");
-};
 
 /**
  * Utility function to get app log path
@@ -74,13 +76,13 @@ export const getServerLogPath = (): string => {
  * @returns {string} The log path.
  */
 export const getAppLogPath = (): string => {
-  const appConfigurations = getAppConfigurations();
+  const appConfigurations = getAppConfigurations()
-  const logFolderPath = join(appConfigurations.data_folder, "logs");
+  const logFolderPath = join(appConfigurations.data_folder, 'logs')
   if (!fs.existsSync(logFolderPath)) {
-    fs.mkdirSync(logFolderPath, { recursive: true });
+    fs.mkdirSync(logFolderPath, { recursive: true })
+  }
+  return join(logFolderPath, 'app.log')
 }
-  return join(logFolderPath, "app.log");
-};
 
 /**
  * Utility function to get data folder path
@@ -88,9 +90,9 @@ export const getAppLogPath = (): string => {
  * @returns {string} The data folder path.
  */
 export const getJanDataFolderPath = (): string => {
-  const appConfigurations = getAppConfigurations();
+  const appConfigurations = getAppConfigurations()
-  return appConfigurations.data_folder;
+  return appConfigurations.data_folder
-};
+}
 
 /**
  * Utility function to get extension path
@@ -98,6 +100,70 @@ export const getJanDataFolderPath = (): string => {
  * @returns {string} The extensions path.
  */
 export const getJanExtensionsPath = (): string => {
-  const appConfigurations = getAppConfigurations();
+  const appConfigurations = getAppConfigurations()
-  return join(appConfigurations.data_folder, "extensions");
+  return join(appConfigurations.data_folder, 'extensions')
-};
+}
+
+/**
+ * Utility function to physical cpu count
+ *
+ * @returns {number} The physical cpu count.
+ */
+export const physicalCpuCount = async (): Promise<number> => {
+  const platform = os.platform()
+  if (platform === 'linux') {
+    const output = await exec('lscpu -p | egrep -v "^#" | sort -u -t, -k 2,4 | wc -l')
+    return parseInt(output.trim(), 10)
+  } else if (platform === 'darwin') {
+    const output = await exec('sysctl -n hw.physicalcpu_max')
+    return parseInt(output.trim(), 10)
+  } else if (platform === 'win32') {
+    const output = await exec('WMIC CPU Get NumberOfCores')
+    return output
+      .split(os.EOL)
+      .map((line: string) => parseInt(line))
+      .filter((value: number) => !isNaN(value))
+      .reduce((sum: number, number: number) => sum + number, 1)
+  } else {
+    const cores = os.cpus().filter((cpu: any, index: number) => {
+      const hasHyperthreading = cpu.model.includes('Intel')
+      const isOdd = index % 2 === 1
+      return !hasHyperthreading || isOdd
+    })
+    return cores.length
+  }
+}
+
+const exec = async (command: string): Promise<string> => {
+  return new Promise((resolve, reject) => {
+    childProcess.exec(command, { encoding: 'utf8' }, (error, stdout) => {
+      if (error) {
+        reject(error)
+      } else {
+        resolve(stdout)
+      }
+    })
+  })
+}
+
+export const getSystemResourceInfo = async (): Promise<SystemResourceInfo> => {
+  const cpu = await physicalCpuCount()
+  const message = `[NITRO]::CPU informations - ${cpu}`
+  log(message)
+  logServer(message)
+
+  return {
+    numCpuPhysicalCore: cpu,
+    memAvailable: 0, // TODO: this should not be 0
+  }
+}
+
+export const getEngineConfiguration = async (engineId: string) => {
+  if (engineId !== 'openai') {
+    return undefined
+  }
+  const directoryPath = join(getJanDataFolderPath(), 'engines')
+  const filePath = join(directoryPath, `${engineId}.json`)
+  const data = fs.readFileSync(filePath, 'utf-8')
+  return JSON.parse(data)
+}

core/src/types/assistant/assistantEntity.ts

@@ -2,6 +2,13 @@
  * Assistant type defines the shape of an assistant object.
  * @stored
  */
+
+export type AssistantTool = {
+  type: string
+  enabled: boolean
+  settings: any
+}
+
 export type Assistant = {
   /** Represents the avatar of the user. */
   avatar: string
@@ -22,7 +29,7 @@ export type Assistant = {
   /** Represents the instructions for the object. */
   instructions?: string
   /** Represents the tools associated with the object. */
-  tools?: any
+  tools?: AssistantTool[]
   /** Represents the file identifiers associated with the object. */
   file_ids: string[]
   /** Represents the metadata of the object. */

core/src/types/index.ts

@@ -6,3 +6,4 @@ export * from './inference'
 export * from './monitoring'
 export * from './file'
 export * from './config'
+export * from './miscellaneous'

core/src/types/inference/inferenceEntity.ts

@@ -1,3 +1,5 @@
+import { ContentType, ContentValue } from '../message'
+
 /**
  * The role of the author of this message.
  */
@@ -13,7 +15,32 @@ export enum ChatCompletionRole {
  */
 export type ChatCompletionMessage = {
   /** The contents of the message. **/
-  content?: string
+  content?: ChatCompletionMessageContent
   /** The role of the author of this message. **/
   role: ChatCompletionRole
 }
+
+export type ChatCompletionMessageContent =
+  | string
+  | (ChatCompletionMessageContentText &
+      ChatCompletionMessageContentImage &
+      ChatCompletionMessageContentDoc)[]
+
+export enum ChatCompletionMessageContentType {
+  Text = 'text',
+  Image = 'image_url',
+  Doc = 'doc_url',
+}
+
+export type ChatCompletionMessageContentText = {
+  type: ChatCompletionMessageContentType
+  text: string
+}
+export type ChatCompletionMessageContentImage = {
+  type: ChatCompletionMessageContentType
+  image_url: { url: string }
+}
+export type ChatCompletionMessageContentDoc = {
+  type: ChatCompletionMessageContentType
+  doc_url: { url: string }
+}

core/src/types/message/messageEntity.ts

@@ -1,5 +1,6 @@
 import { ChatCompletionMessage, ChatCompletionRole } from '../inference'
 import { ModelInfo } from '../model'
+import { Thread } from '../thread'
 
 /**
  * The `ThreadMessage` type defines the shape of a thread's message object.
@@ -35,7 +36,10 @@ export type ThreadMessage = {
 export type MessageRequest = {
   id?: string
 
-  /** The thread id of the message request. **/
+  /**
+   * @deprecated Use thread object instead
+   * The thread id of the message request.
+   */
   threadId: string
 
   /**
@@ -48,6 +52,10 @@ export type MessageRequest = {
 
   /** Settings for constructing a chat completion request **/
   model?: ModelInfo
+
+  /** The thread of this message is belong to. **/
+  // TODO: deprecate threadId field
+  thread?: Thread
 }
 
 /**
@@ -62,7 +70,7 @@ export enum MessageStatus {
   /** Message loaded with error. **/
   Error = 'error',
   /** Message is cancelled streaming */
-  Stopped = "stopped"
+  Stopped = 'stopped',
 }
 
 /**
@@ -71,6 +79,7 @@ export enum MessageStatus {
 export enum ContentType {
   Text = 'text',
   Image = 'image',
+  Pdf = 'pdf',
 }
 
 /**
@@ -80,6 +89,8 @@ export enum ContentType {
 export type ContentValue = {
   value: string
   annotations: string[]
+  name?: string
+  size?: number
 }
 
 /**

core/src/types/miscellaneous/index.ts

@@ -0,0 +1,2 @@
+export * from './systemResourceInfo'
+export * from './promptTemplate'

core/src/types/miscellaneous/promptTemplate.ts

@@ -0,0 +1,6 @@
+export type PromptTemplate = {
+  system_prompt?: string
+  ai_prompt?: string
+  user_prompt?: string
+  error?: string
+}

core/src/types/miscellaneous/systemResourceInfo.ts

@@ -0,0 +1,4 @@
+export type SystemResourceInfo = {
+  numCpuPhysicalCore: number
+  memAvailable: number
+}

core/src/types/model/modelEntity.ts

@@ -7,6 +7,7 @@ export type ModelInfo = {
   settings: ModelSettingParams
   parameters: ModelRuntimeParams
   engine?: InferenceEngine
+  proxyEngine?: InferenceEngine
 }
 
 /**
@@ -18,7 +19,13 @@ export enum InferenceEngine {
   nitro = 'nitro',
   openai = 'openai',
   triton_trtllm = 'triton_trtllm',
-  hf_endpoint = 'hf_endpoint',
+
+  tool_retrieval_enabled = 'tool_retrieval_enabled',
+}
+
+export type ModelArtifact = {
+  filename: string
+  url: string
 }
 
 /**
@@ -45,7 +52,7 @@ export type Model = {
   /**
    * The model download source. It can be an external url or a local filepath.
    */
-  source_url: string
+  sources: ModelArtifact[]
 
   /**
    * The model identifier, which can be referenced in the API endpoints.
@@ -85,6 +92,13 @@ export type Model = {
    * The model engine.
    */
   engine: InferenceEngine
+
+  proxyEngine?: InferenceEngine
+
+  /**
+   * Is multimodal or not.
+   */
+  visionModel?: boolean
 }
 
 export type ModelMetadata = {
@@ -107,6 +121,9 @@ export type ModelSettingParams = {
   system_prompt?: string
   ai_prompt?: string
   user_prompt?: string
+  llama_model_path?: string
+  mmproj?: string
+  cont_batching?: boolean
 }
 
 /**
@@ -122,4 +139,5 @@ export type ModelRuntimeParams = {
   stop?: string[]
   frequency_penalty?: number
   presence_penalty?: number
+  engine?: string
 }

core/src/types/thread/index.ts

@@ -1,2 +1,3 @@
 export * from './threadEntity'
 export * from './threadInterface'
+export * from './threadEvent'

core/src/types/thread/threadEntity.ts

@@ -1,3 +1,4 @@
+import { AssistantTool } from '../assistant'
 import { ModelInfo } from '../model'
 
 /**
@@ -30,6 +31,7 @@ export type ThreadAssistantInfo = {
   assistant_name: string
   model: ModelInfo
   instructions?: string
+  tools?: AssistantTool[]
 }
 
 /**
@@ -41,5 +43,4 @@ export type ThreadState = {
   waitingForResponse: boolean
   error?: Error
   lastMessage?: string
-  isFinishInit?: boolean
 }

core/src/types/thread/threadEvent.ts

@@ -0,0 +1,4 @@
+export enum ThreadEvent {
+  /** The `OnThreadStarted` event is emitted when a thread is started. */
+  OnThreadStarted = 'OnThreadStarted',
+}

docs/.env.example

@@ -1,5 +1,5 @@
 GTM_ID=xxxx
-POSTHOG_PROJECT_API_KEY=xxxx
+UMAMI_PROJECT_API_KEY=xxxx
-POSTHOG_APP_URL=xxxx
+UMAMI_APP_URL=xxxx
 ALGOLIA_API_KEY=xxxx
 ALGOLIA_APP_ID=xxxx

docs/docs/docs/03-engineering/models.md

@@ -56,7 +56,6 @@ jan/                               # Jan root folder
 
 - Each `model` folder contains a `model.json` file, which is a representation of a model.
 - `model.json` contains metadata and default parameters used to run a model.
-- The only required field is `source_url`.
 
 ### Example
 
@@ -66,24 +65,30 @@ Here's a standard example `model.json` for a GGUF model.
 {
   "id": "zephyr-7b",        // Defaults to foldername
   "object": "model",        // Defaults to "model"
-"source_url": "https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/blob/main/zephyr-7b-beta.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "zephyr-7b-beta.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/blob/main/zephyr-7b-beta.Q4_K_M.gguf"
+    }
+  ],
   "name": "Zephyr 7B",      // Defaults to foldername
   "owned_by": "you",        // Defaults to "you"
   "version": "1",           // Defaults to 1
   "created": 1231231,       // Defaults to file creation time
   "description": null,      // Defaults to null
-"state": enum[null, "ready"]
   "format": "ggufv3",       // Defaults to "ggufv3"
   "engine": "nitro",        // engine_id specified in jan/engine folder
-"engine_parameters": {              // Engine parameters inside model.json can override
+  "engine_parameters": {
-    "ctx_len": 2048,                // the value inside the base engine.json
+    // Engine parameters inside model.json can override
+    "ctx_len": 4096,        // the value inside the base engine.json
     "ngl": 100,
     "embedding": true,
-    "n_parallel": 4,
+    "n_parallel": 4
   },
-"model_parameters": {                     // Models are called parameters
+  "model_parameters": {
+    // Models are called parameters
     "stream": true,
-    "max_tokens": 2048,
+    "max_tokens": 4096,
     "stop": ["<endofstring>"], // This usually can be left blank, only used with specific need from model author
     "frequency_penalty": 0,
     "presence_penalty": 0,
@@ -91,8 +96,9 @@ Here's a standard example `model.json` for a GGUF model.
     "top_p": 0.95
   },
   "metadata": {},           // Defaults to {}
-"assets": [                         // Defaults to current dir
+  "assets": [
-    "file://.../zephyr-7b-q4_k_m.bin",
+    // Defaults to current dir
+    "file://.../zephyr-7b-q4_k_m.bin"
   ]
 }
 ```

docs/docs/guides/04-using-models/02-import-manually.mdx

@@ -31,7 +31,6 @@ In this section, we will show you how to import a GGUF model from [HuggingFace](
 
 ## Manually Importing a Downloaded Model (nightly versions and v0.4.4+)
 
-
 ### 1. Create a Model Folder
 
 Navigate to the `~/jan/models` folder. You can find this folder by going to `App Settings` > `Advanced` > `Open App Directory`.
@@ -190,14 +189,18 @@ Edit `model.json` and include the following configurations:
 - Ensure the filename must be `model.json`.
 - Ensure the `id` property matches the folder name you created.
 - Ensure the GGUF filename should match the `id` property exactly.
-- Ensure the `source_url` property is the direct binary download link ending in `.gguf`. In HuggingFace, you can find the direct links in the `Files and versions` tab.
+- Ensure the `source.url` property is the direct binary download link ending in `.gguf`. In HuggingFace, you can find the direct links in the `Files and versions` tab.
 - Ensure you are using the correct `prompt_template`. This is usually provided in the HuggingFace model's description page.
-- Ensure the `state` property is set to `ready`.
 
 ```json title="model.json"
 {
   // highlight-start
-  "source_url": "https://huggingface.co/janhq/trinity-v1-GGUF/resolve/main/trinity-v1.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "trinity-v1.Q4_K_M.gguf",
+      "url": "https://huggingface.co/janhq/trinity-v1-GGUF/resolve/main/trinity-v1.Q4_K_M.gguf"
+    }
+  ],
   "id": "trinity-v1-7b",
   // highlight-end
   "object": "model",
@@ -208,7 +211,8 @@ Edit `model.json` and include the following configurations:
   "settings": {
     "ctx_len": 4096,
     // highlight-next-line
-    "prompt_template": "{system_message}\n### Instruction:\n{prompt}\n### Response:"
+    "prompt_template": "{system_message}\n### Instruction:\n{prompt}\n### Response:",
+    "llama_model_path": "trinity-v1.Q4_K_M.gguf"
   },
   "parameters": {
     "max_tokens": 4096
@@ -218,9 +222,7 @@ Edit `model.json` and include the following configurations:
     "tags": ["7B", "Merged"],
     "size": 4370000000
   },
-  "engine": "nitro",
+  "engine": "nitro"
-  // highlight-next-line
-  "state": "ready"
 }
 ```
 

docs/docs/guides/04-using-models/03-integrate-with-remote-server.mdx

@@ -40,7 +40,12 @@ Navigate to the `~/jan/models` folder. Create a folder named `gpt-3.5-turbo-16k`
 
 ```json title="~/jan/models/gpt-3.5-turbo-16k/model.json"
 {
-  "source_url": "https://openai.com",
+  "sources": [
+    {
+      "filename": "openai",
+      "url": "https://openai.com"
+    }
+  ],
   // highlight-next-line
   "id": "gpt-3.5-turbo-16k",
   "object": "model",
@@ -55,8 +60,7 @@ Navigate to the `~/jan/models` folder. Create a folder named `gpt-3.5-turbo-16k`
     "author": "OpenAI",
     "tags": ["General", "Big Context Length"]
   },
-  "engine": "openai",
+  "engine": "openai"
-  "state": "ready"
   // highlight-end
 }
 ```
@@ -118,7 +122,12 @@ Navigate to the `~/jan/models` folder. Create a folder named `mistral-ins-7b-q4`
 
 ```json title="~/jan/models/mistral-ins-7b-q4/model.json"
 {
-  "source_url": "https://jan.ai",
+  "sources": [
+    {
+      "filename": "janai",
+      "url": "https://jan.ai"
+    }
+  ],
   // highlight-next-line
   "id": "mistral-ins-7b-q4",
   "object": "model",
@@ -134,8 +143,7 @@ Navigate to the `~/jan/models` folder. Create a folder named `mistral-ins-7b-q4`
     "tags": ["remote", "awesome"]
   },
   // highlight-start
-  "engine": "openai",
+  "engine": "openai"
-  "state": "ready"
   // highlight-end
 }
 ```

docs/docs/guides/07-integrations/02-integrate-openrouter.mdx

@@ -49,7 +49,12 @@ Navigate to the `~/jan/models` folder. Create a folder named `<openrouter-modeln
 
 ```json title="~/jan/models/openrouter-dolphin-mixtral-8x7b/model.json"
 {
-  "source_url": "https://openrouter.ai/",
+  "sources": [
+    {
+      "filename": "openrouter",
+      "url": "https://openrouter.ai/"
+    }
+  ],
   "id": "cognitivecomputations/dolphin-mixtral-8x7b",
   "object": "model",
   "name": "Dolphin 2.6 Mixtral 8x7B",
@@ -63,8 +68,7 @@ Navigate to the `~/jan/models` folder. Create a folder named `<openrouter-modeln
     "tags": ["General", "Big Context Length"]
   },
   // highlight-start
-  "engine": "openai",
+  "engine": "openai"
-  "state": "ready"
   // highlight-end
 }
 ```

docs/docs/guides/07-integrations/03-integrate-azure-openai-service.mdx

@@ -59,7 +59,12 @@ Navigate to the `~/jan/models` folder. Create a folder named `<your-deployment-n
 
 ```json title="~/jan/models/gpt-35-hieu-jan/model.json"
 {
-  "source_url": "https://hieujan.openai.azure.com",
+  "sources": [
+    {
+      "filename": "azure_openai",
+      "url": "https://hieujan.openai.azure.com"
+    }
+  ],
   // highlight-next-line
   "id": "gpt-35-hieu-jan",
   "object": "model",
@@ -75,8 +80,7 @@ Navigate to the `~/jan/models` folder. Create a folder named `<your-deployment-n
     "tags": ["General", "Big Context Length"]
   },
   // highlight-start
-  "engine": "openai",
+  "engine": "openai"
-  "state": "ready"
   // highlight-end
 }
 ```

docs/openapi/specs/models.yaml

@@ -59,18 +59,26 @@ components:
           type: string
           description: State format of the model, distinct from the engine.
           example: ggufv3
-        source_url:
+        source:
-          type: string
+          type: array
+          items:
+            type: object
+            properties:
+              url:
                 format: uri
                 description: URL to the source of the model.
                 example: https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/blob/main/zephyr-7b-beta.Q4_K_M.gguf
+              filename:
+                type: string
+                description: Filename of the model.
+                example: zephyr-7b-beta.Q4_K_M.gguf
         settings:
           type: object
           properties:
             ctx_len:
               type: string
               description: Context length.
-              example: "2048"
+              example: "4096"
             ngl:
               type: string
               description: Number of layers.
@@ -94,7 +102,7 @@ components:
             token_limit:
               type: string
               description: Token limit for the model.
-              example: "2048"
+              example: "4096"
             top_k:
               type: string
               description: Top-k setting for the model.
@@ -117,7 +125,7 @@ components:
             type: string
           description: List of assets related to the model.
       required:
-        - source_url
+        - source
     ModelObject:
       type: object
       properties:
@@ -169,11 +177,19 @@ components:
             - running
             - stopped
           description: The current state of the model.
-        source_url:
+        source:
-          type: string
+          type: array
+          items:
+            type: object
+            properties:
+              url:
                 format: uri
                 description: URL to the source of the model.
                 example: https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/blob/main/zephyr-7b-beta.Q4_K_M.gguf
+              filename:
+                type: string
+                description: Filename of the model.
+                example: zephyr-7b-beta.Q4_K_M.gguf
         engine_parameters:
           type: object
           properties:
@@ -208,8 +224,8 @@ components:
                 model.
               minimum: 128
               maximum: 4096
-              default: 2048
+              default: 4096
-              example: 2048
+              example: 4096
             n_parallel:
               type: integer
               description: Number of parallel operations, relevant when continuous batching is
@@ -241,8 +257,8 @@ components:
               description: Maximum context length the model can handle.
               minimum: 0
               maximum: 4096
-              default: 2048
+              default: 4096
-              example: 2048
+              example: 4096
             ngl:
               type: integer
               description: Number of layers in the neural network.
@@ -276,8 +292,8 @@ components:
                 response.
               minimum: 1
               maximum: 4096
-              default: 2048
+              default: 4096
-              example: 2048
+              example: 4096
             top_k:
               type: integer
               description: Limits the model to consider only the top k most likely next tokens
@@ -318,7 +334,7 @@ components:
         - created
         - owned_by
         - state
-        - source_url
+        - source
         - parameters
         - metadata
     DeleteModelResponse:

electron/handlers/app.ts

@@ -1,5 +1,5 @@
 import { app, ipcMain, dialog, shell } from 'electron'
-import { join, basename } from 'path'
+import { join, basename, relative as getRelative, isAbsolute } from 'path'
 import { WindowManager } from './../managers/window'
 import { getResourcePath } from './../utils/path'
 import { AppRoute, AppConfiguration } from '@janhq/core'
@@ -50,6 +50,27 @@ export function handleAppIPCs() {
     join(...paths)
   )
 
+  /**
+   * Checks if the given path is a subdirectory of the given directory.
+   *
+   * @param _event - The IPC event object.
+   * @param from - The path to check.
+   * @param to - The directory to check against.
+   *
+   * @returns {Promise<boolean>} - A promise that resolves with the result.
+   */
+  ipcMain.handle(
+    AppRoute.isSubdirectory,
+    async (_event, from: string, to: string) => {
+      const relative = getRelative(from, to)
+      const isSubdir =
+        relative && !relative.startsWith('..') && !isAbsolute(relative)
+
+      if (isSubdir === '') return false
+      else return isSubdir
+    }
+  )
+
   /**
    * Retrieve basename from given path, respect to the current OS.
    */

electron/handlers/fileManager.ts

@@ -1,4 +1,4 @@
-import { ipcMain } from 'electron'
+import { ipcMain, app } from 'electron'
 // @ts-ignore
 import reflect from '@alumna/reflect'
 
@@ -38,6 +38,10 @@ export function handleFileMangerIPCs() {
     getResourcePath()
   )
 
+  ipcMain.handle(FileManagerRoute.getUserHomePath, async (_event) =>
+    app.getPath('home')
+  )
+
   // handle fs is directory here
   ipcMain.handle(
     FileManagerRoute.fileStat,
@@ -59,4 +63,20 @@ export function handleFileMangerIPCs() {
       return fileStat
     }
   )
+
+  ipcMain.handle(
+    FileManagerRoute.writeBlob,
+    async (_event, path: string, data: string): Promise<void> => {
+      try {
+        const normalizedPath = normalizeFilePath(path)
+        const dataBuffer = Buffer.from(data, 'base64')
+        fs.writeFileSync(
+          join(getJanDataFolderPath(), normalizedPath),
+          dataBuffer
+        )
+      } catch (err) {
+        console.error(`writeFile ${path} result: ${err}`)
+      }
+    }
+  )
 }

electron/handlers/fs.ts

@@ -1,9 +1,9 @@
 import { ipcMain } from 'electron'
 
-import { FileSystemRoute } from '@janhq/core'
-import { join } from 'path'
 import { getJanDataFolderPath, normalizeFilePath } from '@janhq/core/node'
-
+import fs from 'fs'
+import { FileManagerRoute, FileSystemRoute } from '@janhq/core'
+import { join } from 'path'
 /**
  * Handles file system operations.
  */
@@ -15,7 +15,7 @@ export function handleFsIPCs() {
         mdl[route](
           ...args.map((arg) =>
             typeof arg === 'string' &&
-            (arg.includes(`file:/`) || arg.includes(`file:\\`))
+            (arg.startsWith(`file:/`) || arg.startsWith(`file:\\`))
               ? join(getJanDataFolderPath(), normalizeFilePath(arg))
               : arg
           )

electron/main.ts

@@ -28,6 +28,22 @@ import { setupCore } from './utils/setup'
 
 app
   .whenReady()
+  .then(async () => {
+    if (!app.isPackaged) {
+      // Which means you're running from source code
+      const { default: installExtension, REACT_DEVELOPER_TOOLS } = await import(
+        'electron-devtools-installer'
+      ) // Don't use import on top level, since the installer package is dev-only
+      try {
+        const name = installExtension(REACT_DEVELOPER_TOOLS)
+        console.log(`Added Extension: ${name}`)
+      } catch (err) {
+        console.log('An error occurred while installing devtools:')
+        console.error(err)
+        // Only log the error and don't throw it because it's not critical
+      }
+    }
+  })
   .then(setupCore)
   .then(createUserSpace)
   .then(migrateExtensions)

electron/package.json

@@ -86,7 +86,7 @@
     "request": "^2.88.2",
     "request-progress": "^3.0.0",
     "rimraf": "^5.0.5",
-    "typescript": "^5.3.3",
+    "typescript": "^5.2.2",
     "ulid": "^2.3.0",
     "use-debounce": "^9.0.4"
   },
@@ -99,6 +99,7 @@
     "@typescript-eslint/parser": "^6.7.3",
     "electron": "28.0.0",
     "electron-builder": "^24.9.1",
+    "electron-devtools-installer": "^3.2.0",
     "electron-playwright-helpers": "^1.6.0",
     "eslint-plugin-react": "^7.33.2",
     "run-script-os": "^1.1.6"

electron/playwright.config.ts

@@ -1,9 +1,9 @@
-import { PlaywrightTestConfig } from "@playwright/test";
+import { PlaywrightTestConfig } from '@playwright/test'
 
 const config: PlaywrightTestConfig = {
-  testDir: "./tests",
+  testDir: './tests',
   retries: 0,
-  timeout: 120000,
+  globalTimeout: 300000,
-};
+}
 
-export default config;
+export default config

electron/tests/hub.e2e.spec.ts

@@ -9,6 +9,7 @@ import {
 
 let electronApp: ElectronApplication
 let page: Page
+const TIMEOUT: number = parseInt(process.env.TEST_TIMEOUT || '300000')
 
 test.beforeAll(async () => {
   process.env.CI = 'e2e'
@@ -26,7 +27,9 @@ test.beforeAll(async () => {
   })
   await stubDialog(electronApp, 'showMessageBox', { response: 1 })
 
-  page = await electronApp.firstWindow()
+  page = await electronApp.firstWindow({
+    timeout: TIMEOUT,
+  })
 })
 
 test.afterAll(async () => {
@@ -34,8 +37,12 @@ test.afterAll(async () => {
   await page.close()
 })
 
-test('explores models', async () => {
+test('explores hub', async () => {
-  await page.getByTestId('Hub').first().click()
+  test.setTimeout(TIMEOUT)
-  await page.getByTestId('testid-explore-models').isVisible()
+  await page.getByTestId('Hub').first().click({
-  //   More test cases here...
+    timeout: TIMEOUT,
+  })
+  await page.getByTestId('hub-container-test-id').isVisible({
+    timeout: TIMEOUT,
+  })
 })

electron/tests/main.e2e.spec.ts

@@ -1,55 +0,0 @@
-import { _electron as electron } from 'playwright'
-import { ElectronApplication, Page, expect, test } from '@playwright/test'
-
-import {
-  findLatestBuild,
-  parseElectronApp,
-  stubDialog,
-} from 'electron-playwright-helpers'
-
-let electronApp: ElectronApplication
-let page: Page
-
-test.beforeAll(async () => {
-  process.env.CI = 'e2e'
-
-  const latestBuild = findLatestBuild('dist')
-  expect(latestBuild).toBeTruthy()
-
-  // parse the packaged Electron app and find paths and other info
-  const appInfo = parseElectronApp(latestBuild)
-  expect(appInfo).toBeTruthy()
-  expect(appInfo.asar).toBe(true)
-  expect(appInfo.executable).toBeTruthy()
-  expect(appInfo.main).toBeTruthy()
-  expect(appInfo.name).toBe('jan')
-  expect(appInfo.packageJson).toBeTruthy()
-  expect(appInfo.packageJson.name).toBe('jan')
-  expect(appInfo.platform).toBeTruthy()
-  expect(appInfo.platform).toBe(process.platform)
-  expect(appInfo.resourcesDir).toBeTruthy()
-
-  electronApp = await electron.launch({
-    args: [appInfo.main], // main file from package.json
-    executablePath: appInfo.executable, // path to the Electron executable
-  })
-  await stubDialog(electronApp, 'showMessageBox', { response: 1 })
-
-  page = await electronApp.firstWindow()
-})
-
-test.afterAll(async () => {
-  await electronApp.close()
-  await page.close()
-})
-
-test('renders the home page', async () => {
-  expect(page).toBeDefined()
-
-  // Welcome text is available
-  const welcomeText = await page
-    .getByTestId('testid-welcome-title')
-    .first()
-    .isVisible()
-  expect(welcomeText).toBe(false)
-})

electron/tests/navigation.e2e.spec.ts

@@ -9,6 +9,7 @@ import {
 
 let electronApp: ElectronApplication
 let page: Page
+const TIMEOUT: number = parseInt(process.env.TEST_TIMEOUT || '300000')
 
 test.beforeAll(async () => {
   process.env.CI = 'e2e'
@@ -26,7 +27,9 @@ test.beforeAll(async () => {
   })
   await stubDialog(electronApp, 'showMessageBox', { response: 1 })
 
-  page = await electronApp.firstWindow()
+  page = await electronApp.firstWindow({
+    timeout: TIMEOUT,
+  })
 })
 
 test.afterAll(async () => {
@@ -35,20 +38,24 @@ test.afterAll(async () => {
 })
 
 test('renders left navigation panel', async () => {
-  // Chat section should be there
+  test.setTimeout(TIMEOUT)
-  const chatSection = await page.getByTestId('Chat').first().isVisible()
-  expect(chatSection).toBe(false)
-
-  // Home actions
-  /* Disable unstable feature tests
-   ** const botBtn = await page.getByTestId("Bot").first().isEnabled();
-   ** Enable back when it is whitelisted
-   */
-
   const systemMonitorBtn = await page
     .getByTestId('System Monitor')
     .first()
-    .isEnabled()
+    .isEnabled({
-  const settingsBtn = await page.getByTestId('Settings').first().isEnabled()
+      timeout: TIMEOUT,
-  expect([systemMonitorBtn, settingsBtn].filter((e) => !e).length).toBe(0)
+    })
+  const settingsBtn = await page
+    .getByTestId('Thread')
+    .first()
+    .isEnabled({ timeout: TIMEOUT })
+  expect([systemMonitorBtn, settingsBtn].filter((e) => !e).length).toBe(0)
+  // Chat section should be there
+  await page.getByTestId('Local API Server').first().click({
+    timeout: TIMEOUT,
+  })
+  const localServer = await page.getByTestId('local-server-testid').first()
+  await expect(localServer).toBeVisible({
+    timeout: TIMEOUT,
+  })
 })

electron/tests/settings.e2e.spec.ts

@@ -9,6 +9,7 @@ import {
 
 let electronApp: ElectronApplication
 let page: Page
+const TIMEOUT: number = parseInt(process.env.TEST_TIMEOUT || '300000')
 
 test.beforeAll(async () => {
   process.env.CI = 'e2e'
@@ -26,7 +27,9 @@ test.beforeAll(async () => {
   })
   await stubDialog(electronApp, 'showMessageBox', { response: 1 })
 
-  page = await electronApp.firstWindow()
+  page = await electronApp.firstWindow({
+    timeout: TIMEOUT,
+  })
 })
 
 test.afterAll(async () => {
@@ -35,6 +38,8 @@ test.afterAll(async () => {
 })
 
 test('shows settings', async () => {
-  await page.getByTestId('Settings').first().click()
+  test.setTimeout(TIMEOUT)
-  await page.getByTestId('testid-setting-description').isVisible()
+  await page.getByTestId('Settings').first().click({ timeout: TIMEOUT })
+  const settingDescription = page.getByTestId('testid-setting-description')
+  await expect(settingDescription).toBeVisible({ timeout: TIMEOUT })
 })

electron/tests/system-monitor.e2e.spec.ts

@@ -1,41 +0,0 @@
-import { _electron as electron } from 'playwright'
-import { ElectronApplication, Page, expect, test } from '@playwright/test'
-
-import {
-  findLatestBuild,
-  parseElectronApp,
-  stubDialog,
-} from 'electron-playwright-helpers'
-
-let electronApp: ElectronApplication
-let page: Page
-
-test.beforeAll(async () => {
-  process.env.CI = 'e2e'
-
-  const latestBuild = findLatestBuild('dist')
-  expect(latestBuild).toBeTruthy()
-
-  // parse the packaged Electron app and find paths and other info
-  const appInfo = parseElectronApp(latestBuild)
-  expect(appInfo).toBeTruthy()
-
-  electronApp = await electron.launch({
-    args: [appInfo.main], // main file from package.json
-    executablePath: appInfo.executable, // path to the Electron executable
-  })
-  await stubDialog(electronApp, 'showMessageBox', { response: 1 })
-
-  page = await electronApp.firstWindow()
-})
-
-test.afterAll(async () => {
-  await electronApp.close()
-  await page.close()
-})
-
-test('shows system monitor', async () => {
-  await page.getByTestId('System Monitor').first().click()
-  await page.getByTestId('testid-system-monitor').isVisible()
-  //   More test cases here...
-})

extensions/assistant-extension/package.json

@@ -3,26 +3,50 @@
   "version": "1.0.0",
   "description": "This extension enables assistants, including Jan, a default assistant that can call all downloaded models",
   "main": "dist/index.js",
-  "module": "dist/module.js",
+  "node": "dist/node/index.js",
   "author": "Jan <service@jan.ai>",
   "license": "AGPL-3.0",
   "scripts": {
-    "build": "tsc -b . && webpack --config webpack.config.js",
+    "build": "tsc --module commonjs && rollup -c rollup.config.ts",
-    "build:publish": "rimraf *.tgz --glob && npm run build && npm pack && cpx *.tgz ../../electron/pre-install"
+    "build:publish:linux": "rimraf *.tgz --glob && npm run build && npm pack && cpx *.tgz ../../electron/pre-install",
+    "build:publish:darwin": "rimraf *.tgz --glob && npm run build && ../../.github/scripts/auto-sign.sh && npm pack && cpx *.tgz ../../electron/pre-install",
+    "build:publish:win32": "rimraf *.tgz --glob && npm run build && npm pack && cpx *.tgz ../../electron/pre-install",
+    "build:publish": "run-script-os"
   },
   "devDependencies": {
+    "@rollup/plugin-commonjs": "^25.0.7",
+    "@rollup/plugin-json": "^6.1.0",
+    "@rollup/plugin-node-resolve": "^15.2.3",
+    "@rollup/plugin-replace": "^5.0.5",
+    "@types/pdf-parse": "^1.1.4",
+    "cpx": "^1.5.0",
     "rimraf": "^3.0.2",
-    "webpack": "^5.88.2",
+    "rollup": "^2.38.5",
-    "webpack-cli": "^5.1.4"
+    "rollup-plugin-define": "^1.0.1",
+    "rollup-plugin-sourcemaps": "^0.6.3",
+    "rollup-plugin-typescript2": "^0.36.0",
+    "typescript": "^5.3.3",
+    "run-script-os": "^1.1.6"
   },
   "dependencies": {
     "@janhq/core": "file:../../core",
+    "@langchain/community": "0.0.13",
+    "hnswlib-node": "^1.4.2",
+    "langchain": "^0.0.214",
     "path-browserify": "^1.0.1",
+    "pdf-parse": "^1.1.1",
     "ts-loader": "^9.5.0"
   },
   "files": [
     "dist/*",
     "package.json",
     "README.md"
+  ],
+  "bundleDependencies": [
+    "@janhq/core",
+    "@langchain/community",
+    "hnswlib-node",
+    "langchain",
+    "pdf-parse"
   ]
 }

extensions/assistant-extension/rollup.config.ts

@@ -0,0 +1,81 @@
+import resolve from "@rollup/plugin-node-resolve";
+import commonjs from "@rollup/plugin-commonjs";
+import sourceMaps from "rollup-plugin-sourcemaps";
+import typescript from "rollup-plugin-typescript2";
+import json from "@rollup/plugin-json";
+import replace from "@rollup/plugin-replace";
+
+const packageJson = require("./package.json");
+
+const pkg = require("./package.json");
+
+export default [
+  {
+    input: `src/index.ts`,
+    output: [{ file: pkg.main, format: "es", sourcemap: true }],
+    // Indicate here external modules you don't wanna include in your bundle (i.e.: 'lodash')
+    external: [],
+    watch: {
+      include: "src/**",
+    },
+    plugins: [
+      replace({
+        NODE: JSON.stringify(`${packageJson.name}/${packageJson.node}`),
+        EXTENSION_NAME: JSON.stringify(packageJson.name),
+        VERSION: JSON.stringify(packageJson.version),
+      }),
+      // Allow json resolution
+      json(),
+      //     Compile TypeScript files
+      typescript({ useTsconfigDeclarationDir: true }),
+      // Compile TypeScript files
+      // Allow bundling cjs modules (unlike webpack, rollup doesn't understand cjs)
+      commonjs(),
+      // Allow node_modules resolution, so you can use 'external' to control
+      // which external modules to include in the bundle
+      // https://github.com/rollup/rollup-plugin-node-resolve#usage
+      resolve({
+        extensions: [".js", ".ts", ".svelte"],
+      }),
+
+      // Resolve source maps to the original source
+      sourceMaps(),
+    ],
+  },
+  {
+    input: `src/node/index.ts`,
+    output: [{ dir: "dist/node", format: "cjs", sourcemap: false }],
+    // Indicate here external modules you don't wanna include in your bundle (i.e.: 'lodash')
+    external: [
+      "@janhq/core/node",
+      "@langchain/community",
+      "langchain",
+      "langsmith",
+      "path",
+      "hnswlib-node",
+    ],
+    watch: {
+      include: "src/node/**",
+    },
+    // inlineDynamicImports: true,
+    plugins: [
+      // Allow json resolution
+      json(),
+      // Compile TypeScript files
+      typescript({ useTsconfigDeclarationDir: true }),
+      // Allow bundling cjs modules (unlike webpack, rollup doesn't understand cjs)
+      commonjs({
+        ignoreDynamicRequires: true,
+      }),
+      // Allow node_modules resolution, so you can use 'external' to control
+      // which external modules to include in the bundle
+      // https://github.com/rollup/rollup-plugin-node-resolve#usage
+      resolve({
+        extensions: [".ts", ".js", ".json"],
+      }),
+
+      // Resolve source maps to the original source
+      // sourceMaps(),
+    ],
+  },
+];

extensions/assistant-extension/src/@types/global.d.ts

@@ -1 +1,3 @@
-declare const MODULE: string;
+declare const NODE: string;
+declare const EXTENSION_NAME: string;
+declare const VERSION: string;

extensions/assistant-extension/src/index.ts

@@ -1,33 +1,174 @@
-import { fs, Assistant } from "@janhq/core";
+import {
-import { AssistantExtension } from "@janhq/core";
+  fs,
-import { join } from "path";
+  Assistant,
+  MessageRequest,
+  events,
+  InferenceEngine,
+  MessageEvent,
+  InferenceEvent,
+  joinPath,
+  executeOnMain,
+  AssistantExtension,
+} from "@janhq/core";
 
 export default class JanAssistantExtension extends AssistantExtension {
   private static readonly _homeDir = "file://assistants";
 
+  controller = new AbortController();
+  isCancelled = false;
+  retrievalThreadId: string | undefined = undefined;
+
   async onLoad() {
     // making the assistant directory
-    if (!(await fs.existsSync(JanAssistantExtension._homeDir)))
+    const assistantDirExist = await fs.existsSync(
-      fs.mkdirSync(JanAssistantExtension._homeDir).then(() => {
+      JanAssistantExtension._homeDir,
+    );
+    if (
+      localStorage.getItem(`${EXTENSION_NAME}-version`) !== VERSION ||
+      !assistantDirExist
+    ) {
+      if (!assistantDirExist)
+        await fs.mkdirSync(JanAssistantExtension._homeDir);
+
+      // Write assistant metadata
       this.createJanAssistant();
+      // Finished migration
+      localStorage.setItem(`${EXTENSION_NAME}-version`, VERSION);
+    }
+
+    // Events subscription
+    events.on(MessageEvent.OnMessageSent, (data: MessageRequest) =>
+      JanAssistantExtension.handleMessageRequest(data, this),
+    );
+
+    events.on(InferenceEvent.OnInferenceStopped, () => {
+      JanAssistantExtension.handleInferenceStopped(this);
     });
   }
 
+  private static async handleInferenceStopped(instance: JanAssistantExtension) {
+    instance.isCancelled = true;
+    instance.controller?.abort();
+  }
+
+  private static async handleMessageRequest(
+    data: MessageRequest,
+    instance: JanAssistantExtension,
+  ) {
+    instance.isCancelled = false;
+    instance.controller = new AbortController();
+
+    if (
+      data.model?.engine !== InferenceEngine.tool_retrieval_enabled ||
+      !data.messages ||
+      !data.thread?.assistants[0]?.tools
+    ) {
+      return;
+    }
+
+    const latestMessage = data.messages[data.messages.length - 1];
+
+    // Ingest the document if needed
+    if (
+      latestMessage &&
+      latestMessage.content &&
+      typeof latestMessage.content !== "string"
+    ) {
+      const docFile = latestMessage.content[1]?.doc_url?.url;
+      if (docFile) {
+        await executeOnMain(
+          NODE,
+          "toolRetrievalIngestNewDocument",
+          docFile,
+          data.model?.proxyEngine,
+        );
+      }
+    }
+
+    // Load agent on thread changed
+    if (instance.retrievalThreadId !== data.threadId) {
+      await executeOnMain(NODE, "toolRetrievalLoadThreadMemory", data.threadId);
+
+      instance.retrievalThreadId = data.threadId;
+
+      // Update the text splitter
+      await executeOnMain(
+        NODE,
+        "toolRetrievalUpdateTextSplitter",
+        data.thread.assistants[0].tools[0]?.settings?.chunk_size ?? 4000,
+        data.thread.assistants[0].tools[0]?.settings?.chunk_overlap ?? 200,
+      );
+    }
+
+    if (latestMessage.content) {
+      const prompt =
+        typeof latestMessage.content === "string"
+          ? latestMessage.content
+          : latestMessage.content[0].text;
+      // Retrieve the result
+      console.debug("toolRetrievalQuery", latestMessage.content);
+      const retrievalResult = await executeOnMain(
+        NODE,
+        "toolRetrievalQueryResult",
+        prompt,
+      );
+
+      // Update the message content
+      // Using the retrieval template with the result and query
+      if (data.thread?.assistants[0].tools)
+        data.messages[data.messages.length - 1].content =
+          data.thread.assistants[0].tools[0].settings?.retrieval_template
+            ?.replace("{CONTEXT}", retrievalResult)
+            .replace("{QUESTION}", prompt);
+    }
+
+    // Filter out all the messages that are not text
+    data.messages = data.messages.map((message) => {
+      if (
+        message.content &&
+        typeof message.content !== "string" &&
+        (message.content.length ?? 0) > 0
+      ) {
+        return {
+          ...message,
+          content: [message.content[0]],
+        };
+      }
+      return message;
+    });
+
+    // Reroute the result to inference engine
+    const output = {
+      ...data,
+      model: {
+        ...data.model,
+        engine: data.model.proxyEngine,
+      },
+    };
+    events.emit(MessageEvent.OnMessageSent, output);
+  }
+
   /**
    * Called when the extension is unloaded.
    */
   onUnload(): void {}
 
   async createAssistant(assistant: Assistant): Promise<void> {
-    const assistantDir = join(JanAssistantExtension._homeDir, assistant.id);
+    const assistantDir = await joinPath([
+      JanAssistantExtension._homeDir,
+      assistant.id,
+    ]);
     if (!(await fs.existsSync(assistantDir))) await fs.mkdirSync(assistantDir);
 
     // store the assistant metadata json
-    const assistantMetadataPath = join(assistantDir, "assistant.json");
+    const assistantMetadataPath = await joinPath([
+      assistantDir,
+      "assistant.json",
+    ]);
     try {
       await fs.writeFileSync(
         assistantMetadataPath,
-        JSON.stringify(assistant, null, 2)
+        JSON.stringify(assistant, null, 2),
       );
     } catch (err) {
       console.error(err);
@@ -39,14 +180,17 @@ export default class JanAssistantExtension extends AssistantExtension {
     // get all the assistant metadata json
     const results: Assistant[] = [];
     const allFileName: string[] = await fs.readdirSync(
-      JanAssistantExtension._homeDir
+      JanAssistantExtension._homeDir,
     );
     for (const fileName of allFileName) {
-      const filePath = join(JanAssistantExtension._homeDir, fileName);
+      const filePath = await joinPath([
+        JanAssistantExtension._homeDir,
+        fileName,
+      ]);
 
       if (filePath.includes(".DS_Store")) continue;
       const jsonFiles: string[] = (await fs.readdirSync(filePath)).filter(
-        (file: string) => file === "assistant.json"
+        (file: string) => file === "assistant.json",
       );
 
       if (jsonFiles.length !== 1) {
@@ -55,8 +199,8 @@ export default class JanAssistantExtension extends AssistantExtension {
       }
 
       const content = await fs.readFileSync(
-        join(filePath, jsonFiles[0]),
+        await joinPath([filePath, jsonFiles[0]]),
-        "utf-8"
+        "utf-8",
       );
       const assistant: Assistant =
         typeof content === "object" ? content : JSON.parse(content);
@@ -73,7 +217,10 @@ export default class JanAssistantExtension extends AssistantExtension {
     }
 
     // remove the directory
-    const assistantDir = join(JanAssistantExtension._homeDir, assistant.id);
+    const assistantDir = await joinPath([
+      JanAssistantExtension._homeDir,
+      assistant.id,
+    ]);
     await fs.rmdirSync(assistantDir);
     return Promise.resolve();
   }
@@ -89,7 +236,24 @@ export default class JanAssistantExtension extends AssistantExtension {
       description: "A default assistant that can use all downloaded models",
       model: "*",
       instructions: "",
-      tools: undefined,
+      tools: [
+        {
+          type: "retrieval",
+          enabled: false,
+          settings: {
+            top_k: 2,
+            chunk_size: 1024,
+            chunk_overlap: 64,
+            retrieval_template: `Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.
+            ----------------
+            CONTEXT: {CONTEXT}
+            ----------------
+            QUESTION: {QUESTION}
+            ----------------
+            Helpful Answer:`,
+          },
+        },
+      ],
       file_ids: [],
       metadata: undefined,
     };

extensions/assistant-extension/src/node/engine.ts

@@ -0,0 +1,13 @@
+import fs from "fs";
+import path from "path";
+import { getJanDataFolderPath } from "@janhq/core/node";
+
+// Sec: Do not send engine settings over requests
+// Read it manually instead
+export const readEmbeddingEngine = (engineName: string) => {
+  const engineSettings = fs.readFileSync(
+    path.join(getJanDataFolderPath(), "engines", `${engineName}.json`),
+    "utf-8",
+  );
+  return JSON.parse(engineSettings);
+};

extensions/assistant-extension/src/node/index.ts

@@ -0,0 +1,39 @@
+import { getJanDataFolderPath, normalizeFilePath } from "@janhq/core/node";
+import { Retrieval } from "./tools/retrieval";
+import path from "path";
+
+const retrieval = new Retrieval();
+
+export async function toolRetrievalUpdateTextSplitter(
+  chunkSize: number,
+  chunkOverlap: number,
+) {
+  retrieval.updateTextSplitter(chunkSize, chunkOverlap);
+  return Promise.resolve();
+}
+export async function toolRetrievalIngestNewDocument(
+  file: string,
+  engine: string,
+) {
+  const filePath = path.join(getJanDataFolderPath(), normalizeFilePath(file));
+  const threadPath = path.dirname(filePath.replace("files", ""));
+  retrieval.updateEmbeddingEngine(engine);
+  await retrieval.ingestAgentKnowledge(filePath, `${threadPath}/memory`);
+  return Promise.resolve();
+}
+
+export async function toolRetrievalLoadThreadMemory(threadId: string) {
+  try {
+    await retrieval.loadRetrievalAgent(
+      path.join(getJanDataFolderPath(), "threads", threadId, "memory"),
+    );
+    return Promise.resolve();
+  } catch (err) {
+    console.debug(err);
+  }
+}
+
+export async function toolRetrievalQueryResult(query: string) {
+  const res = await retrieval.generateResult(query);
+  return Promise.resolve(res);
+}

extensions/assistant-extension/src/node/tools/retrieval/index.ts

@@ -0,0 +1,79 @@
+import { RecursiveCharacterTextSplitter } from "langchain/text_splitter";
+import { formatDocumentsAsString } from "langchain/util/document";
+import { PDFLoader } from "langchain/document_loaders/fs/pdf";
+
+import { HNSWLib } from "langchain/vectorstores/hnswlib";
+
+import { OpenAIEmbeddings } from "langchain/embeddings/openai";
+import { readEmbeddingEngine } from "../../engine";
+
+export class Retrieval {
+  public chunkSize: number = 100;
+  public chunkOverlap?: number = 0;
+  private retriever: any;
+
+  private embeddingModel?: OpenAIEmbeddings = undefined;
+  private textSplitter?: RecursiveCharacterTextSplitter;
+
+  constructor(chunkSize: number = 4000, chunkOverlap: number = 200) {
+    this.updateTextSplitter(chunkSize, chunkOverlap);
+  }
+
+  public updateTextSplitter(chunkSize: number, chunkOverlap: number): void {
+    this.chunkSize = chunkSize;
+    this.chunkOverlap = chunkOverlap;
+    this.textSplitter = new RecursiveCharacterTextSplitter({
+      chunkSize: chunkSize,
+      chunkOverlap: chunkOverlap,
+    });
+  }
+
+  public updateEmbeddingEngine(engine: string): void {
+    // Engine settings are not compatible with the current embedding model params
+    // Switch case manually for now
+    const settings = readEmbeddingEngine(engine);
+    if (engine === "nitro") {
+      this.embeddingModel = new OpenAIEmbeddings(
+        { openAIApiKey: "nitro-embedding" },
+        { basePath: "http://127.0.0.1:3928/v1" }
+      );
+    } else {
+      // Fallback to OpenAI Settings
+      this.embeddingModel = new OpenAIEmbeddings({
+        configuration: {
+          apiKey: settings.api_key,
+        },
+      });
+    }
+  }
+
+  public ingestAgentKnowledge = async (
+    filePath: string,
+    memoryPath: string
+  ): Promise<any> => {
+    const loader = new PDFLoader(filePath, {
+      splitPages: true,
+    });
+    if (!this.embeddingModel) return Promise.reject();
+    const doc = await loader.load();
+    const docs = await this.textSplitter!.splitDocuments(doc);
+    const vectorStore = await HNSWLib.fromDocuments(docs, this.embeddingModel);
+    return vectorStore.save(memoryPath);
+  };
+
+  public loadRetrievalAgent = async (memoryPath: string): Promise<void> => {
+    if (!this.embeddingModel) return Promise.reject();
+    const vectorStore = await HNSWLib.load(memoryPath, this.embeddingModel);
+    this.retriever = vectorStore.asRetriever(2);
+    return Promise.resolve();
+  };
+
+  public generateResult = async (query: string): Promise<string> => {
+    if (!this.retriever) {
+      return Promise.resolve(" ");
+    }
+    const relevantDocs = await this.retriever.getRelevantDocuments(query);
+    const serializedDoc = formatDocumentsAsString(relevantDocs);
+    return Promise.resolve(serializedDoc);
+  };
+}

extensions/assistant-extension/tsconfig.json

@@ -1,14 +1,20 @@
 {
   "compilerOptions": {
-    "target": "es2016",
-    "module": "ES6",
     "moduleResolution": "node",
-    "outDir": "./dist",
+    "target": "es5",
-    "esModuleInterop": true,
+    "module": "ES2020",
-    "forceConsistentCasingInFileNames": true,
+    "lib": ["es2015", "es2016", "es2017", "dom"],
-    "strict": false,
+    "strict": true,
+    "sourceMap": true,
+    "declaration": true,
+    "allowSyntheticDefaultImports": true,
+    "experimentalDecorators": true,
+    "emitDecoratorMetadata": true,
+    "declarationDir": "dist/types",
+    "outDir": "dist",
+    "importHelpers": true,
+    "typeRoots": ["node_modules/@types"],
     "skipLibCheck": true,
-    "rootDir": "./src"
   },
-  "include": ["./src"]
+  "include": ["src"],
 }

extensions/assistant-extension/webpack.config.js

@@ -1,38 +0,0 @@
-const path = require("path");
-const webpack = require("webpack");
-const packageJson = require("./package.json");
-
-module.exports = {
-  experiments: { outputModule: true },
-  entry: "./src/index.ts", // Adjust the entry point to match your project's main file
-  mode: "production",
-  module: {
-    rules: [
-      {
-        test: /\.tsx?$/,
-        use: "ts-loader",
-        exclude: /node_modules/,
-      },
-    ],
-  },
-  output: {
-    filename: "index.js", // Adjust the output file name as needed
-    path: path.resolve(__dirname, "dist"),
-    library: { type: "module" }, // Specify ESM output format
-  },
-  plugins: [
-    new webpack.DefinePlugin({
-      MODULE: JSON.stringify(`${packageJson.name}/${packageJson.module}`),
-    }),
-  ],
-  resolve: {
-    extensions: [".ts", ".js"],
-    fallback: {
-      path: require.resolve("path-browserify"),
-    },
-  },
-  optimization: {
-    minimize: false,
-  },
-  // Add loaders and other configuration as needed for your project
-};

extensions/conversational-extension/src/index.ts

@@ -4,15 +4,14 @@ import {
   ConversationalExtension,
   Thread,
   ThreadMessage,
+  events,
 } from '@janhq/core'
 
 /**
  * JSONConversationalExtension is a ConversationalExtension implementation that provides
  * functionality for managing threads.
  */
-export default class JSONConversationalExtension
+export default class JSONConversationalExtension extends ConversationalExtension {
-  extends ConversationalExtension
-{
   private static readonly _homeDir = 'file://threads'
   private static readonly _threadInfoFileName = 'thread.json'
   private static readonly _threadMessagesFileName = 'messages.jsonl'
@@ -119,6 +118,33 @@ export default class JSONConversationalExtension
       ])
       if (!(await fs.existsSync(threadDirPath)))
         await fs.mkdirSync(threadDirPath)
+
+      if (message.content[0]?.type === 'image') {
+        const filesPath = await joinPath([threadDirPath, 'files'])
+        if (!(await fs.existsSync(filesPath))) await fs.mkdirSync(filesPath)
+
+        const imagePath = await joinPath([filesPath, `${message.id}.png`])
+        const base64 = message.content[0].text.annotations[0]
+        await this.storeImage(base64, imagePath)
+        if ((await fs.existsSync(imagePath)) && message.content?.length) {
+          // Use file path instead of blob
+          message.content[0].text.annotations[0] = `threads/${message.thread_id}/files/${message.id}.png`
+        }
+      }
+
+      if (message.content[0]?.type === 'pdf') {
+        const filesPath = await joinPath([threadDirPath, 'files'])
+        if (!(await fs.existsSync(filesPath))) await fs.mkdirSync(filesPath)
+
+        const filePath = await joinPath([filesPath, `${message.id}.pdf`])
+        const blob = message.content[0].text.annotations[0]
+        await this.storeFile(blob, filePath)
+
+        if ((await fs.existsSync(filePath)) && message.content?.length) {
+          // Use file path instead of blob
+          message.content[0].text.annotations[0] = `threads/${message.thread_id}/files/${message.id}.pdf`
+        }
+      }
       await fs.appendFileSync(threadMessagePath, JSON.stringify(message) + '\n')
       Promise.resolve()
     } catch (err) {
@@ -126,6 +152,25 @@ export default class JSONConversationalExtension
     }
   }
 
+  async storeImage(base64: string, filePath: string): Promise<void> {
+    const base64Data = base64.replace(/^data:image\/\w+;base64,/, '')
+
+    try {
+      await fs.writeBlob(filePath, base64Data)
+    } catch (err) {
+      console.error(err)
+    }
+  }
+
+  async storeFile(base64: string, filePath: string): Promise<void> {
+    const base64Data = base64.replace(/^data:application\/pdf;base64,/, '')
+    try {
+      await fs.writeBlob(filePath, base64Data)
+    } catch (err) {
+      console.error(err)
+    }
+  }
+
   async writeMessages(
     threadId: string,
     messages: ThreadMessage[]
@@ -229,7 +274,11 @@ export default class JSONConversationalExtension
 
       const messages: ThreadMessage[] = []
       result.forEach((line: string) => {
+        try {
           messages.push(JSON.parse(line) as ThreadMessage)
+        } catch (err) {
+          console.error(err)
+        }
       })
       return messages
     } catch (err) {

extensions/inference-nitro-extension/bin/version.txt

@@ -1 +1 @@
-0.2.12
+0.3.5

extensions/inference-nitro-extension/package.json

@@ -35,11 +35,12 @@
     "rollup-plugin-sourcemaps": "^0.6.3",
     "rollup-plugin-typescript2": "^0.36.0",
     "run-script-os": "^1.1.6",
-    "typescript": "^5.3.3"
+    "typescript": "^5.2.2"
   },
   "dependencies": {
     "@janhq/core": "file:../../core",
     "@rollup/plugin-replace": "^5.0.5",
+    "@types/os-utils": "^0.0.4",
     "fetch-retry": "^5.0.6",
     "path-browserify": "^1.0.1",
     "rxjs": "^7.8.1",

extensions/inference-nitro-extension/src/@types/global.d.ts

@@ -2,22 +2,6 @@ declare const NODE: string;
 declare const INFERENCE_URL: string;
 declare const TROUBLESHOOTING_URL: string;
 
-/**
- * The parameters for the initModel function.
- * @property settings - The settings for the machine learning model.
- * @property settings.ctx_len - The context length.
- * @property settings.ngl - The number of generated tokens.
- * @property settings.cont_batching - Whether to use continuous batching.
- * @property settings.embedding - Whether to use embedding.
- */
-interface EngineSettings {
-  ctx_len: number;
-  ngl: number;
-  cpu_threads: number;
-  cont_batching: boolean;
-  embedding: boolean;
-}
-
 /**
  * The response from the initModel function.
  * @property error - An error message if the model fails to load.
@@ -26,8 +10,3 @@ interface ModelOperationResponse {
   error?: any;
   modelFile?: string;
 }
-
-interface ResourcesInfo {
-  numCpuPhysicalCore: number;
-  memAvailable: number;
-}

extensions/inference-nitro-extension/src/index.ts

@@ -24,6 +24,7 @@ import {
   MessageEvent,
   ModelEvent,
   InferenceEvent,
+  ModelSettingParams,
 } from "@janhq/core";
 import { requestInference } from "./helpers/sse";
 import { ulid } from "ulid";
@@ -45,12 +46,12 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
 
   private _currentModel: Model | undefined;
 
-  private _engineSettings: EngineSettings = {
+  private _engineSettings: ModelSettingParams = {
     ctx_len: 2048,
     ngl: 100,
     cpu_threads: 1,
     cont_batching: false,
-    embedding: false,
+    embedding: true,
   };
 
   controller = new AbortController();
@@ -83,19 +84,19 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
 
     // Events subscription
     events.on(MessageEvent.OnMessageSent, (data: MessageRequest) =>
-      this.onMessageRequest(data)
+      this.onMessageRequest(data),
     );
 
     events.on(ModelEvent.OnModelInit, (model: Model) =>
-      this.onModelInit(model)
+      this.onModelInit(model),
     );
 
     events.on(ModelEvent.OnModelStop, (model: Model) =>
-      this.onModelStop(model)
+      this.onModelStop(model),
     );
 
     events.on(InferenceEvent.OnInferenceStopped, () =>
-      this.onInferenceStopped()
+      this.onInferenceStopped(),
     );
 
     // Attempt to fetch nvidia info
@@ -120,7 +121,7 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
       } else {
         await fs.writeFileSync(
           engineFile,
-          JSON.stringify(this._engineSettings, null, 2)
+          JSON.stringify(this._engineSettings, null, 2),
         );
       }
     } catch (err) {
@@ -133,6 +134,7 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
 
     const modelFullPath = await joinPath(["models", model.id]);
 
+    this._currentModel = model;
     const nitroInitResult = await executeOnMain(NODE, "runModel", {
       modelFullPath,
       model,
@@ -143,12 +145,11 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
       return;
     }
 
-    this._currentModel = model;
     events.emit(ModelEvent.OnModelReady, model);
 
     this.getNitroProcesHealthIntervalId = setInterval(
       () => this.periodicallyGetNitroHealth(),
-      JanInferenceNitroExtension._intervalHealthCheck
+      JanInferenceNitroExtension._intervalHealthCheck,
     );
   }
 
@@ -225,6 +226,9 @@ export default class JanInferenceNitroExtension extends InferenceExtension {
    */
   private async onMessageRequest(data: MessageRequest) {
     if (data.model?.engine !== InferenceEngine.nitro || !this._currentModel) {
+      console.log(
+        `Model is not nitro or no model loaded ${data.model?.engine} ${this._currentModel}`
+      );
       return;
     }
 

extensions/inference-nitro-extension/src/node/index.ts

@@ -3,11 +3,19 @@ import path from "path";
 import { ChildProcessWithoutNullStreams, spawn } from "child_process";
 import tcpPortUsed from "tcp-port-used";
 import fetchRT from "fetch-retry";
-import { log, getJanDataFolderPath } from "@janhq/core/node";
+import {
+  log,
+  getJanDataFolderPath,
+  getSystemResourceInfo,
+} from "@janhq/core/node";
 import { getNitroProcessInfo, updateNvidiaInfo } from "./nvidia";
-import { Model, InferenceEngine, ModelSettingParams } from "@janhq/core";
+import {
+  Model,
+  InferenceEngine,
+  ModelSettingParams,
+  PromptTemplate,
+} from "@janhq/core";
 import { executableNitroFile } from "./execute";
-import { physicalCpuCount } from "./utils";
 
 // Polyfill fetch with retry
 const fetchRetry = fetchRT(fetch);
@@ -19,25 +27,6 @@ interface ModelInitOptions {
   modelFullPath: string;
   model: Model;
 }
-
-/**
- * The response object of Prompt Template parsing.
- */
-interface PromptTemplate {
-  system_prompt?: string;
-  ai_prompt?: string;
-  user_prompt?: string;
-  error?: string;
-}
-
-/**
- * Model setting args for Nitro model load.
- */
-interface ModelSettingArgs extends ModelSettingParams {
-  llama_model_path: string;
-  cpu_threads: number;
-}
-
 // The PORT to use for the Nitro subprocess
 const PORT = 3928;
 // The HOST address to use for the Nitro subprocess
@@ -60,7 +49,7 @@ let subprocess: ChildProcessWithoutNullStreams | undefined = undefined;
 // The current model file url
 let currentModelFile: string = "";
 // The current model settings
-let currentSettings: ModelSettingArgs | undefined = undefined;
+let currentSettings: ModelSettingParams | undefined = undefined;
 
 /**
  * Stops a Nitro subprocess.
@@ -78,7 +67,7 @@ function stopModel(): Promise<void> {
  * TODO: Should pass absolute of the model file instead of just the name - So we can modurize the module.ts to npm package
  */
 async function runModel(
-  wrapper: ModelInitOptions
+  wrapper: ModelInitOptions,
 ): Promise<ModelOperationResponse | void> {
   if (wrapper.model.engine !== InferenceEngine.nitro) {
     // Not a nitro model
@@ -96,7 +85,7 @@ async function runModel(
   const ggufBinFile = files.find(
     (file) =>
       file === path.basename(currentModelFile) ||
-      file.toLowerCase().includes(SUPPORTED_MODEL_FORMAT)
+      file.toLowerCase().includes(SUPPORTED_MODEL_FORMAT),
   );
 
   if (!ggufBinFile) return Promise.reject("No GGUF model file found");
@@ -106,7 +95,7 @@ async function runModel(
   if (wrapper.model.engine !== InferenceEngine.nitro) {
     return Promise.reject("Not a nitro model");
   } else {
-    const nitroResourceProbe = await getResourcesInfo();
+    const nitroResourceProbe = await getSystemResourceInfo();
     // Convert settings.prompt_template to system_prompt, user_prompt, ai_prompt
     if (wrapper.model.settings.prompt_template) {
       const promptTemplate = wrapper.model.settings.prompt_template;
@@ -119,13 +108,20 @@ async function runModel(
       wrapper.model.settings.ai_prompt = prompt.ai_prompt;
     }
 
+    const modelFolderPath = path.join(janRoot, "models", wrapper.model.id);
+    const modelPath = wrapper.model.settings.llama_model_path
+      ? path.join(modelFolderPath, wrapper.model.settings.llama_model_path)
+      : currentModelFile;
+
     currentSettings = {
-      llama_model_path: currentModelFile,
       ...wrapper.model.settings,
+      llama_model_path: modelPath,
       // This is critical and requires real CPU physical core count (or performance core)
       cpu_threads: Math.max(1, nitroResourceProbe.numCpuPhysicalCore),
+      ...(wrapper.model.settings.mmproj && {
+        mmproj: path.join(modelFolderPath, wrapper.model.settings.mmproj),
+      }),
     };
-    console.log(currentSettings);
     return runNitroAndLoadModel();
   }
 }
@@ -184,10 +180,10 @@ function promptTemplateConverter(promptTemplate: string): PromptTemplate {
     const system_prompt = promptTemplate.substring(0, systemIndex);
     const user_prompt = promptTemplate.substring(
       systemIndex + systemMarker.length,
-      promptIndex
+      promptIndex,
     );
     const ai_prompt = promptTemplate.substring(
-      promptIndex + promptMarker.length
+      promptIndex + promptMarker.length,
     );
 
     // Return the split parts
@@ -197,7 +193,7 @@ function promptTemplateConverter(promptTemplate: string): PromptTemplate {
     const promptIndex = promptTemplate.indexOf(promptMarker);
     const user_prompt = promptTemplate.substring(0, promptIndex);
     const ai_prompt = promptTemplate.substring(
-      promptIndex + promptMarker.length
+      promptIndex + promptMarker.length,
     );
 
     // Return the split parts
@@ -213,6 +209,9 @@ function promptTemplateConverter(promptTemplate: string): PromptTemplate {
  * @returns A Promise that resolves when the model is loaded successfully, or rejects with an error message if the model is not found or fails to load.
  */
 function loadLLMModel(settings: any): Promise<Response> {
+  if (!settings?.ngl) {
+    settings.ngl = 100;
+  }
   log(`[NITRO]::Debug: Loading model with params ${JSON.stringify(settings)}`);
   return fetchRetry(NITRO_HTTP_LOAD_MODEL_URL, {
     method: "POST",
@@ -226,14 +225,14 @@ function loadLLMModel(settings: any): Promise<Response> {
     .then((res) => {
       log(
         `[NITRO]::Debug: Load model success with response ${JSON.stringify(
-          res
+          res,
-        )}`
+        )}`,
       );
       return Promise.resolve(res);
     })
     .catch((err) => {
       log(`[NITRO]::Error: Load model failed with error ${err}`);
-      return Promise.reject();
+      return Promise.reject(err);
     });
 }
 
@@ -255,8 +254,8 @@ async function validateModelStatus(): Promise<void> {
     retryDelay: 500,
   }).then(async (res: Response) => {
     log(
-      `[NITRO]::Debug: Validate model state success with response ${JSON.stringify(
+      `[NITRO]::Debug: Validate model state with response ${JSON.stringify(
-        res
+        res.status
       )}`
     );
     // If the response is OK, check model_loaded status.
@@ -265,9 +264,19 @@ async function validateModelStatus(): Promise<void> {
       // If the model is loaded, return an empty object.
       // Otherwise, return an object with an error message.
       if (body.model_loaded) {
+        log(
+          `[NITRO]::Debug: Validate model state success with response ${JSON.stringify(
+            body
+          )}`
+        );
         return Promise.resolve();
       }
     }
+    log(
+      `[NITRO]::Debug: Validate model state failed with response ${JSON.stringify(
+        res.statusText
+      )}`
+    );
     return Promise.reject("Validate model status failed");
   });
 }
@@ -308,7 +317,7 @@ function spawnNitroProcess(): Promise<any> {
     const args: string[] = ["1", LOCAL_HOST, PORT.toString()];
     // Execute the binary
     log(
-      `[NITRO]::Debug: Spawn nitro at path: ${executableOptions.executablePath}, and args: ${args}`
+      `[NITRO]::Debug: Spawn nitro at path: ${executableOptions.executablePath}, and args: ${args}`,
     );
     subprocess = spawn(
       executableOptions.executablePath,
@@ -319,7 +328,7 @@ function spawnNitroProcess(): Promise<any> {
           ...process.env,
           CUDA_VISIBLE_DEVICES: executableOptions.cudaVisibleDevices,
         },
-      }
+      },
     );
 
     // Handle subprocess output
@@ -344,22 +353,6 @@ function spawnNitroProcess(): Promise<any> {
   });
 }
 
-/**
- * Get the system resources information
- * TODO: Move to Core so that it can be reused
- */
-function getResourcesInfo(): Promise<ResourcesInfo> {
-  return new Promise(async (resolve) => {
-    const cpu = await physicalCpuCount();
-    log(`[NITRO]::CPU informations - ${cpu}`);
-    const response: ResourcesInfo = {
-      numCpuPhysicalCore: cpu,
-      memAvailable: 0,
-    };
-    resolve(response);
-  });
-}
-
 /**
  * Every module should have a dispose function
  * This will be called when the extension is unloaded and should clean up any resources

extensions/inference-nitro-extension/src/node/utils.ts

@@ -1,56 +0,0 @@
-import os from "os";
-import childProcess from "child_process";
-
-function exec(command: string): Promise<string> {
-  return new Promise((resolve, reject) => {
-    childProcess.exec(command, { encoding: "utf8" }, (error, stdout) => {
-      if (error) {
-        reject(error);
-      } else {
-        resolve(stdout);
-      }
-    });
-  });
-}
-
-let amount: number;
-const platform = os.platform();
-
-export async function physicalCpuCount(): Promise<number> {
-  return new Promise((resolve, reject) => {
-    if (platform === "linux") {
-      exec('lscpu -p | egrep -v "^#" | sort -u -t, -k 2,4 | wc -l')
-        .then((output) => {
-          amount = parseInt(output.trim(), 10);
-          resolve(amount);
-        })
-        .catch(reject);
-    } else if (platform === "darwin") {
-      exec("sysctl -n hw.physicalcpu_max")
-        .then((output) => {
-          amount = parseInt(output.trim(), 10);
-          resolve(amount);
-        })
-        .catch(reject);
-    } else if (platform === "win32") {
-      exec("WMIC CPU Get NumberOfCores")
-        .then((output) => {
-          amount = output
-            .split(os.EOL)
-            .map((line: string) => parseInt(line))
-            .filter((value: number) => !isNaN(value))
-            .reduce((sum: number, number: number) => sum + number, 1);
-          resolve(amount);
-        })
-        .catch(reject);
-    } else {
-      const cores = os.cpus().filter((cpu: any, index: number) => {
-        const hasHyperthreading = cpu.model.includes("Intel");
-        const isOdd = index % 2 === 1;
-        return !hasHyperthreading || isOdd;
-      });
-      amount = cores.length;
-      resolve(amount);
-    }
-  });
-}

extensions/inference-openai-extension/src/index.ts

@@ -15,6 +15,7 @@ import {
   ThreadMessage,
   events,
   fs,
+  InferenceEngine,
   BaseExtension,
   MessageEvent,
   ModelEvent,
@@ -114,7 +115,7 @@ export default class JanInferenceOpenAIExtension extends BaseExtension {
     }
   }
   private static async handleModelInit(model: OpenAIModel) {
-    if (model.engine !== "openai") {
+    if (model.engine !== InferenceEngine.openai) {
       return;
     } else {
       JanInferenceOpenAIExtension._currentModel = model;

extensions/inference-openai-extension/tsconfig.json

@@ -3,13 +3,12 @@
     "target": "es2016",
     "module": "ES6",
     "moduleResolution": "node",
-
     "outDir": "./dist",
     "esModuleInterop": true,
     "forceConsistentCasingInFileNames": true,
     "strict": false,
     "skipLibCheck": true,
-    "rootDir": "./src"
+    "rootDir": "./src",
   },
-  "include": ["./src"]
+  "include": ["./src"],
 }

extensions/inference-triton-trtllm-extension/tsconfig.json

@@ -3,13 +3,12 @@
     "target": "es2016",
     "module": "ES6",
     "moduleResolution": "node",
-
     "outDir": "./dist",
     "esModuleInterop": true,
     "forceConsistentCasingInFileNames": true,
     "strict": false,
     "skipLibCheck": true,
-    "rootDir": "./src"
+    "rootDir": "./src",
   },
-  "include": ["./src"]
+  "include": ["./src"],
 }

extensions/model-extension/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@janhq/model-extension",
-  "version": "1.0.22",
+  "version": "1.0.23",
   "description": "Model Management Extension provides model exploration and seamless downloads",
   "main": "dist/index.js",
   "module": "dist/module.js",

extensions/model-extension/src/index.ts

@@ -80,16 +80,34 @@ export default class JanModelExtension extends ModelExtension {
     const modelDirPath = await joinPath([JanModelExtension._homeDir, model.id])
     if (!(await fs.existsSync(modelDirPath))) await fs.mkdirSync(modelDirPath)
 
-    // try to retrieve the download file name from the source url
+    if (model.sources.length > 1) {
-    // if it fails, use the model ID as the file name
+      // path to model binaries
-    const extractedFileName = await model.source_url.split('/').pop()
+      for (const source of model.sources) {
+        let path = this.extractFileName(source.url)
+        if (source.filename) {
+          path = await joinPath([modelDirPath, source.filename])
+        }
+
+        downloadFile(source.url, path, network)
+      }
+    } else {
+      const fileName = this.extractFileName(model.sources[0]?.url)
+      const path = await joinPath([modelDirPath, fileName])
+      downloadFile(model.sources[0]?.url, path, network)
+    }
+  }
+
+  /**
+   *  try to retrieve the download file name from the source url
+   */
+  private extractFileName(url: string): string {
+    const extractedFileName = url.split('/').pop()
     const fileName = extractedFileName
       .toLowerCase()
       .endsWith(JanModelExtension._supportedModelFormat)
       ? extractedFileName
-      : model.id
+      : extractedFileName + JanModelExtension._supportedModelFormat
-    const path = await joinPath([modelDirPath, fileName])
+    return fileName
-    downloadFile(model.source_url, path, network)
   }
 
   /**
@@ -98,6 +116,7 @@ export default class JanModelExtension extends ModelExtension {
    * @returns {Promise<void>} A promise that resolves when the download has been cancelled.
    */
   async cancelModelDownload(modelId: string): Promise<void> {
+    const model = await this.getConfiguredModels()
     return abortDownload(
       await joinPath([JanModelExtension._homeDir, modelId, modelId])
     ).then(async () => {
@@ -163,15 +182,16 @@ export default class JanModelExtension extends ModelExtension {
           .then((files: string[]) => {
             // or model binary exists in the directory
             // model binary name can match model ID or be a .gguf file and not be an incompleted model file
+            // TODO: Check diff between urls, filenames
             return (
               files.includes(modelDir) ||
-              files.some(
+              files.filter(
                 (file) =>
                   file
                     .toLowerCase()
                     .includes(JanModelExtension._supportedModelFormat) &&
                   !file.endsWith(JanModelExtension._incompletedModelFileName)
-              )
+              )?.length >= model.sources.length
             )
           })
       }
@@ -198,7 +218,6 @@ export default class JanModelExtension extends ModelExtension {
 
       const readJsonPromises = allDirectories.map(async (dirName) => {
         // filter out directories that don't match the selector
-
         // read model.json
         const jsonPath = await joinPath([
           JanModelExtension._homeDir,
@@ -226,7 +245,21 @@ export default class JanModelExtension extends ModelExtension {
       const modelData = results.map((result) => {
         if (result.status === 'fulfilled') {
           try {
-            return result.value as Model
+            // This to ensure backward compatibility with `model.json` with `source_url`
+            const tmpModel =
+              typeof result.value === 'object'
+                ? result.value
+                : JSON.parse(result.value)
+            if (tmpModel['source_url'] != null) {
+              tmpModel['source'] = [
+                {
+                  filename: tmpModel.id,
+                  url: tmpModel['source_url'],
+                },
+              ]
+            }
+
+            return tmpModel as Model
           } catch {
             console.debug(`Unable to parse model metadata: ${result.value}`)
             return undefined

extensions/monitoring-extension/src/index.ts

@@ -1,5 +1,4 @@
-import { MonitoringExtension } from "@janhq/core";
+import { MonitoringExtension, executeOnMain } from "@janhq/core";
-import { executeOnMain } from "@janhq/core";
 
 /**
  * JanMonitoringExtension is a extension that provides system monitoring functionality.

models/bakllava-1/model.json

@@ -0,0 +1,33 @@
+{
+  "sources": [
+    {
+      "filename": "ggml-model-q5_k.gguf",
+      "url": "https://huggingface.co/mys/ggml_bakllava-1/resolve/main/ggml-model-q5_k.gguf"
+    },
+    {
+      "filename": "mmproj-model-f16.gguf",
+      "url": "https://huggingface.co/mys/ggml_bakllava-1/resolve/main/mmproj-model-f16.gguf"
+    }
+  ],
+  "id": "bakllava-1",
+  "object": "model",
+  "name": "BakLlava 1",
+  "version": "1.0",
+  "description": "BakLlava 1 can bring vision understanding to Jan",
+  "format": "gguf",
+  "settings": {
+    "ctx_len": 4096,
+    "prompt_template": "\n### Instruction:\n{prompt}\n### Response:\n",
+    "llama_model_path": "ggml-model-q5_k.gguf",
+    "mmproj": "mmproj-model-f16.gguf"
+  },
+  "parameters": {
+    "max_tokens": 4096
+  },
+  "metadata": {
+    "author": "Mys",
+    "tags": ["Vision"],
+    "size": 5750000000
+  },
+  "engine": "nitro"
+}

models/capybara-34b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Nous-Capybara-34B-GGUF/resolve/main/nous-capybara-34b.Q5_K_M.gguf",
+  "sources": [
+    {
+      "filename": "nous-capybara-34b.Q5_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Nous-Capybara-34B-GGUF/resolve/main/nous-capybara-34b.Q5_K_M.gguf"
+    }
+  ],
   "id": "capybara-34b",
   "object": "model",
   "name": "Capybara 200k 34B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "USER:\n{prompt}\nASSISTANT:"
+    "prompt_template": "USER:\n{prompt}\nASSISTANT:",
+    "llama_model_path": "nous-capybara-34b.Q5_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/codeninja-1.0-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/beowolx/CodeNinja-1.0-OpenChat-7B-GGUF/resolve/main/codeninja-1.0-openchat-7b.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "codeninja-1.0-openchat-7b.Q4_K_M.gguf",
+      "url": "https://huggingface.co/beowolx/CodeNinja-1.0-OpenChat-7B-GGUF/resolve/main/codeninja-1.0-openchat-7b.Q4_K_M.gguf"
+    }
+  ],
   "id": "codeninja-1.0-7b",
   "object": "model",
   "name": "CodeNinja 7B Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "GPT4 Correct User: {prompt}<|end_of_turn|>GPT4 Correct Assistant:"
+    "prompt_template": "GPT4 Correct User: {prompt}<|end_of_turn|>GPT4 Correct Assistant:",
+    "llama_model_path": "codeninja-1.0-openchat-7b.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": ["<|end_of_turn|>"],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },
@@ -26,4 +31,3 @@
   },
   "engine": "nitro"
 }
-  

models/config/default-model.json

@@ -2,7 +2,12 @@
   "object": "model",
   "version": 1,
   "format": "gguf",
-  "source_url": "N/A",
+  "sources": [
+    {
+      "url": "N/A",
+      "filename": "N/A"
+    }
+  ],
   "id": "N/A",
   "name": "N/A",
   "created": 0,
@@ -10,7 +15,8 @@
   "settings": {
     "ctx_len": 4096,
     "embedding": false,
-    "prompt_template": "{system_message}\n### Instruction: {prompt}\n### Response:"
+    "prompt_template": "{system_message}\n### Instruction: {prompt}\n### Response:",
+    "llama_model_path": "N/A"
   },
   "parameters": {
     "temperature": 0.7,

models/deepseek-coder-1.3b/model.json

@@ -1,6 +1,10 @@
-
 {
-    "source_url": "https://huggingface.co/TheBloke/deepseek-coder-1.3b-instruct-GGUF/resolve/main/deepseek-coder-1.3b-instruct.Q8_0.gguf",
+  "sources": [
+    {
+      "filename": "deepseek-coder-1.3b-instruct.Q8_0.gguf",
+      "url": "https://huggingface.co/TheBloke/deepseek-coder-1.3b-instruct-GGUF/resolve/main/deepseek-coder-1.3b-instruct.Q8_0.gguf"
+    }
+  ],
   "id": "deepseek-coder-1.3b",
   "object": "model",
   "name": "Deepseek Coder 1.3B Q8",
@@ -9,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### Instruction:\n{prompt}\n### Response:"
+    "prompt_template": "### Instruction:\n{prompt}\n### Response:",
+    "llama_model_path": "deepseek-coder-1.3b-instruct.Q8_0.gguf"
   },
   "parameters": {
     "temperature": 0.7,

models/deepseek-coder-34b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/deepseek-coder-33B-instruct-GGUF/resolve/main/deepseek-coder-33b-instruct.Q5_K_M.gguf",
+  "sources": [
+    {
+      "filename": "deepseek-coder-33b-instruct.Q5_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/deepseek-coder-33B-instruct-GGUF/resolve/main/deepseek-coder-33b-instruct.Q5_K_M.gguf"
+    }
+  ],
   "id": "deepseek-coder-34b",
   "object": "model",
   "name": "Deepseek Coder 33B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### Instruction:\n{prompt}\n### Response:"
+    "prompt_template": "### Instruction:\n{prompt}\n### Response:",
+    "llama_model_path": "deepseek-coder-33b-instruct.Q5_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/dolphin-2.7-mixtral-8x7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/dolphin-2.7-mixtral-8x7b-GGUF/resolve/main/dolphin-2.7-mixtral-8x7b.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "dolphin-2.7-mixtral-8x7b.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/dolphin-2.7-mixtral-8x7b-GGUF/resolve/main/dolphin-2.7-mixtral-8x7b.Q4_K_M.gguf"
+    }
+  ],
   "id": "dolphin-2.7-mixtral-8x7b",
   "object": "model",
   "name": "Dolphin 8x7B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
+    "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant",
+    "llama_model_path": "dolphin-2.7-mixtral-8x7b.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,

models/gpt-3.5-turbo-16k-0613/model.json

@@ -1,5 +1,9 @@
 {
-    "source_url": "https://openai.com",
+  "sources": [
+    {
+      "url": "https://openai.com"
+    }
+  ],
   "id": "gpt-3.5-turbo-16k-0613",
   "object": "model",
   "name": "OpenAI GPT 3.5 Turbo 16k 0613",
@@ -12,7 +16,5 @@
     "author": "OpenAI",
     "tags": ["General", "Big Context Length"]
   },
-    "engine": "openai",
+  "engine": "openai"
-    "state": "ready"
 }
-  

models/gpt-3.5-turbo/model.json

@@ -1,5 +1,9 @@
 {
-    "source_url": "https://openai.com",
+  "sources": [
+    {
+      "url": "https://openai.com"
+    }
+  ],
   "id": "gpt-3.5-turbo",
   "object": "model",
   "name": "OpenAI GPT 3.5 Turbo",
@@ -12,7 +16,5 @@
     "author": "OpenAI",
     "tags": ["General", "Big Context Length"]
   },
-    "engine": "openai",
+  "engine": "openai"
-    "state": "ready"
 }
-  

models/gpt-4/model.json

@@ -1,5 +1,9 @@
 {
-    "source_url": "https://openai.com",
+  "sources": [
+    {
+      "url": "https://openai.com"
+    }
+  ],
   "id": "gpt-4",
   "object": "model",
   "name": "OpenAI GPT 4",
@@ -12,7 +16,5 @@
     "author": "OpenAI",
     "tags": ["General", "Big Context Length"]
   },
-    "engine": "openai",
+  "engine": "openai"
-    "state": "ready"
 }
-  

models/llama2-chat-70b-q4/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Llama-2-70B-Chat-GGUF/resolve/main/llama-2-70b-chat.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "llama-2-70b-chat.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Llama-2-70B-Chat-GGUF/resolve/main/llama-2-70b-chat.Q4_K_M.gguf"
+    }
+  ],
   "id": "llama2-chat-70b-q4",
   "object": "model",
   "name": "Llama 2 Chat 70B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "[INST] <<SYS>>\n{system_message}<</SYS>>\n{prompt}[/INST]"
+    "prompt_template": "[INST] <<SYS>>\n{system_message}<</SYS>>\n{prompt}[/INST]",
+    "llama_model_path": "llama-2-70b-chat.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/llama2-chat-7b-q4/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "llama-2-7b-chat.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf"
+    }
+  ],
   "id": "llama2-chat-7b-q4",
   "object": "model",
   "name": "Llama 2 Chat 7B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "[INST] <<SYS>>\n{system_message}<</SYS>>\n{prompt}[/INST]"
+    "prompt_template": "[INST] <<SYS>>\n{system_message}<</SYS>>\n{prompt}[/INST]",
+    "llama_model_path": "llama-2-7b-chat.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/llava-1.5-13b-q5/model.json

@@ -0,0 +1,33 @@
+{
+  "sources": [
+    {
+      "filename": "ggml-model-q5_k.gguf",
+      "url": "https://huggingface.co/mys/ggml_llava-v1.5-13b/resolve/main/ggml-model-q5_k.gguf"
+    },
+    {
+      "filename": "mmproj-model-f16.gguf",
+      "url": "https://huggingface.co/mys/ggml_llava-v1.5-13b/resolve/main/mmproj-model-f16.gguf"
+    }
+  ],
+  "id": "llava-1.5-13b-q5",
+  "object": "model",
+  "name": "LlaVa 1.5 13B Q5 K",
+  "version": "1.0",
+  "description": "LlaVa 1.5 can bring vision understanding to Jan",
+  "format": "gguf",
+  "settings": {
+    "ctx_len": 4096,
+    "prompt_template": "\n### Instruction:\n{prompt}\n### Response:\n",
+    "llama_model_path": "ggml-model-q5_k.gguf",
+    "mmproj": "mmproj-model-f16.gguf"
+  },
+  "parameters": {
+    "max_tokens": 4096
+  },
+  "metadata": {
+    "author": "Mys",
+    "tags": ["Vision"],
+    "size": 9850000000
+  },
+  "engine": "nitro"
+}

models/llava-1.5-7b-q5/model.json

@@ -0,0 +1,33 @@
+{
+  "sources": [
+    {
+      "filename": "ggml-model-q5_k.gguf",
+      "url": "https://huggingface.co/mys/ggml_llava-v1.5-7b/resolve/main/ggml-model-q5_k.gguf"
+    },
+    {
+      "filename": "mmproj-model-f16.gguf",
+      "url": "https://huggingface.co/mys/ggml_llava-v1.5-7b/resolve/main/mmproj-model-f16.gguf"
+    }
+  ],
+  "id": "llava-1.5-7b-q5",
+  "object": "model",
+  "name": "LlaVa 1.5 7B Q5 K",
+  "version": "1.0",
+  "description": "LlaVa 1.5 can bring vision understanding to Jan",
+  "format": "gguf",
+  "settings": {
+    "ctx_len": 4096,
+    "prompt_template": "\n### Instruction:\n{prompt}\n### Response:\n",
+    "llama_model_path": "ggml-model-q5_k.gguf",
+    "mmproj": "mmproj-model-f16.gguf"
+  },
+  "parameters": {
+    "max_tokens": 4096
+  },
+  "metadata": {
+    "author": "Mys",
+    "tags": ["Vision"],
+    "size": 5400000000
+  },
+  "engine": "nitro"
+}

models/mistral-ins-7b-q4/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "mistral-7b-instruct-v0.2.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf"
+    }
+  ],
   "id": "mistral-ins-7b-q4",
   "object": "model",
   "name": "Mistral Instruct 7B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "[INST] {prompt} [/INST]"
+    "prompt_template": "[INST] {prompt} [/INST]",
+    "llama_model_path": "mistral-7b-instruct-v0.2.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -27,4 +33,3 @@
   },
   "engine": "nitro"
 }
-  

models/mixtral-8x7b-instruct/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/resolve/main/mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Mixtral-8x7B-Instruct-v0.1-GGUF/resolve/main/mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf"
+    }
+  ],
   "id": "mixtral-8x7b-instruct",
   "object": "model",
   "name": "Mixtral 8x7B Instruct Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "[INST] {prompt} [/INST]"
+    "prompt_template": "[INST] {prompt} [/INST]",
+    "llama_model_path": "mixtral-8x7b-instruct-v0.1.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },

models/noromaid-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/NeverSleep/Noromaid-7b-v0.1.1-GGUF/resolve/main/Noromaid-7b-v0.1.1.q5_k_m.gguf",
+  "sources": [
+    {
+      "filename": "Noromaid-7b-v0.1.1.q5_k_m.gguf",
+      "url": "https://huggingface.co/NeverSleep/Noromaid-7b-v0.1.1-GGUF/resolve/main/Noromaid-7b-v0.1.1.q5_k_m.gguf"
+    }
+  ],
   "id": "noromaid-7b",
   "object": "model",
   "name": "Noromaid 7B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### Instruction:{prompt}\n### Response:"
+    "prompt_template": "### Instruction:{prompt}\n### Response:",
+    "llama_model_path": "Noromaid-7b-v0.1.1.q5_k_m.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/openchat-3.5-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/openchat-3.5-1210-GGUF/resolve/main/openchat-3.5-1210.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "openchat-3.5-1210.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/openchat-3.5-1210-GGUF/resolve/main/openchat-3.5-1210.Q4_K_M.gguf"
+    }
+  ],
   "id": "openchat-3.5-7b",
   "object": "model",
   "name": "Openchat-3.5 7B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "GPT4 Correct User: {prompt}<|end_of_turn|>GPT4 Correct Assistant:"
+    "prompt_template": "GPT4 Correct User: {prompt}<|end_of_turn|>GPT4 Correct Assistant:",
+    "llama_model_path": "openchat-3.5-1210.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,

models/openhermes-neural-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/janhq/openhermes-2.5-neural-chat-v3-3-slerp-GGUF/resolve/main/openhermes-2.5-neural-chat-v3-3-slerp.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "openhermes-2.5-neural-chat-v3-3-slerp.Q4_K_M.gguf",
+      "url": "https://huggingface.co/janhq/openhermes-2.5-neural-chat-v3-3-slerp-GGUF/resolve/main/openhermes-2.5-neural-chat-v3-3-slerp.Q4_K_M.gguf"
+    }
+  ],
   "id": "openhermes-neural-7b",
   "object": "model",
   "name": "OpenHermes Neural 7B Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
+    "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant",
+    "llama_model_path": "openhermes-2.5-neural-chat-v3-3-slerp.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },

models/phi-2-3b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/phi-2-GGUF/resolve/main/phi-2.Q8_0.gguf",
+  "sources": [
+    {
+      "filename": "phi-2.Q8_0.gguf",
+      "url": "https://huggingface.co/TheBloke/phi-2-GGUF/resolve/main/phi-2.Q8_0.gguf"
+    }
+  ],
   "id": "phi-2-3b",
   "object": "model",
   "name": "Phi-2 3B Q8",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "Intruct:\n{prompt}\nOutput:"
+    "prompt_template": "Intruct:\n{prompt}\nOutput:",
+    "llama_model_path": "phi-2.Q8_0.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/phind-34b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Phind-CodeLlama-34B-v2-GGUF/resolve/main/phind-codellama-34b-v2.Q5_K_M.gguf",
+  "sources": [
+    {
+      "filename": "phind-codellama-34b-v2.Q5_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Phind-CodeLlama-34B-v2-GGUF/resolve/main/phind-codellama-34b-v2.Q5_K_M.gguf"
+    }
+  ],
   "id": "phind-34b",
   "object": "model",
   "name": "Phind 34B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### System Prompt\n{system_message}\n### User Message\n{prompt}\n### Assistant"
+    "prompt_template": "### System Prompt\n{system_message}\n### User Message\n{prompt}\n### Assistant",
+    "llama_model_path": "phind-codellama-34b-v2.Q5_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/solar-10.7b-slerp/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/janhq/Solar-10.7B-SLERP-GGUF/resolve/main/solar-10.7b-slerp.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "solar-10.7b-slerp.Q4_K_M.gguf",
+      "url": "https://huggingface.co/janhq/Solar-10.7B-SLERP-GGUF/resolve/main/solar-10.7b-slerp.Q4_K_M.gguf"
+    }
+  ],
   "id": "solar-10.7b-slerp",
   "object": "model",
   "name": "Solar Slerp 10.7B Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### User: {prompt}\n### Assistant:"
+    "prompt_template": "### User: {prompt}\n### Assistant:",
+    "llama_model_path": "solar-10.7b-slerp.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },
@@ -26,4 +31,3 @@
   },
   "engine": "nitro"
 }
-  

models/starling-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/resolve/main/starling-lm-7b-alpha.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "starling-lm-7b-alpha.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Starling-LM-7B-alpha-GGUF/resolve/main/starling-lm-7b-alpha.Q4_K_M.gguf"
+    }
+  ],
   "id": "starling-7b",
   "object": "model",
   "name": "Starling alpha 7B Q4",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:"
+    "prompt_template": "GPT4 User: {prompt}<|end_of_turn|>GPT4 Assistant:",
+    "llama_model_path": "starling-lm-7b-alpha.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/stealth-v1.2-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/janhq/stealth-v1.3-GGUF/resolve/main/stealth-v1.3.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "stealth-v1.3.Q4_K_M.gguf",
+      "url": "https://huggingface.co/janhq/stealth-v1.3-GGUF/resolve/main/stealth-v1.3.Q4_K_M.gguf"
+    }
+  ],
   "id": "stealth-v1.2-7b",
   "object": "model",
   "name": "Stealth 7B Q4",
@@ -8,24 +13,20 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
+    "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant",
+    "llama_model_path": "stealth-v1.3.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },
   "metadata": {
     "author": "Jan",
-      "tags": [
+    "tags": ["7B", "Finetuned", "Featured"],
-        "7B",
-        "Finetuned",
-        "Featured"
-      ],
     "size": 4370000000
   },
   "engine": "nitro"

models/tinyllama-1.1b/model.json

@@ -1,5 +1,10 @@
 {
-  "source_url": "https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"
+    }
+  ],
   "id": "tinyllama-1.1b",
   "object": "model",
   "name": "TinyLlama Chat 1.1B Q4",
@@ -7,8 +12,9 @@
   "description": "TinyLlama is a tiny model with only 1.1B. It's a good model for less powerful computers.",
   "format": "gguf",
   "settings": {
-      "ctx_len": 2048,
+    "ctx_len": 4096,
-      "prompt_template": "<|system|>\n{system_message}<|user|>\n{prompt}<|assistant|>"
+    "prompt_template": "<|system|>\n{system_message}<|user|>\n{prompt}<|assistant|>",
+    "llama_model_path": "tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,

models/trinity-v1.2-7b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/janhq/trinity-v1.2-GGUF/resolve/main/trinity-v1.2.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "trinity-v1.2.Q4_K_M.gguf",
+      "url": "https://huggingface.co/janhq/trinity-v1.2-GGUF/resolve/main/trinity-v1.2.Q4_K_M.gguf"
+    }
+  ],
   "id": "trinity-v1.2-7b",
   "object": "model",
   "name": "Trinity-v1.2 7B Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
+    "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant",
+    "llama_model_path": "trinity-v1.2.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },

models/tulu-2-70b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/tulu-2-dpo-70B-GGUF/resolve/main/tulu-2-dpo-70b.Q4_K_M.gguf",
+  "sources": [
+    {
+      "filename": "tulu-2-dpo-70b.Q4_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/tulu-2-dpo-70B-GGUF/resolve/main/tulu-2-dpo-70b.Q4_K_M.gguf"
+    }
+  ],
   "id": "tulu-2-70b",
   "object": "model",
   "name": "Tulu 2 70B Q4",
@@ -8,14 +13,14 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|user|>\n{prompt}\n<|assistant|>"
+    "prompt_template": "<|user|>\n{prompt}\n<|assistant|>",
+    "llama_model_path": "tulu-2-dpo-70b.Q4_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },

models/wizardcoder-13b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/WizardCoder-Python-13B-V1.0-GGUF/resolve/main/wizardcoder-python-13b-v1.0.Q5_K_M.gguf",
+  "sources": [
+    {
+      "filename": "wizardcoder-python-13b-v1.0.Q5_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/WizardCoder-Python-13B-V1.0-GGUF/resolve/main/wizardcoder-python-13b-v1.0.Q5_K_M.gguf"
+    }
+  ],
   "id": "wizardcoder-13b",
   "object": "model",
   "name": "Wizard Coder Python 13B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "### Instruction:\n{prompt}\n### Response:"
+    "prompt_template": "### Instruction:\n{prompt}\n### Response:",
+    "llama_model_path": "wizardcoder-python-13b-v1.0.Q5_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-  

models/yarn-mistral-7b/model.json

@@ -1,5 +1,9 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Yarn-Mistral-7B-128k-GGUF/resolve/main/yarn-mistral-7b-128k.Q4_K_M.gguf",
+  "sources": [
+    {
+      "url": "https://huggingface.co/TheBloke/Yarn-Mistral-7B-128k-GGUF/resolve/main/yarn-mistral-7b-128k.Q4_K_M.gguf"
+    }
+  ],
   "id": "yarn-mistral-7b",
   "object": "model",
   "name": "Yarn Mistral 7B Q4",
@@ -15,7 +19,6 @@
     "top_p": 0.95,
     "stream": true,
     "max_tokens": 4096,
-      "stop": [],
     "frequency_penalty": 0,
     "presence_penalty": 0
   },
@@ -26,4 +29,3 @@
   },
   "engine": "nitro"
 }
-  

models/yi-34b/model.json

@@ -1,5 +1,10 @@
 {
-    "source_url": "https://huggingface.co/TheBloke/Yi-34B-Chat-GGUF/resolve/main/yi-34b-chat.Q5_K_M.gguf",
+  "sources": [
+    {
+      "filename": "yi-34b-chat.Q5_K_M.gguf",
+      "url": "https://huggingface.co/TheBloke/Yi-34B-Chat-GGUF/resolve/main/yi-34b-chat.Q5_K_M.gguf"
+    }
+  ],
   "id": "yi-34b",
   "object": "model",
   "name": "Yi 34B Q5",
@@ -8,7 +13,8 @@
   "format": "gguf",
   "settings": {
     "ctx_len": 4096,
-      "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
+    "prompt_template": "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant",
+    "llama_model_path": "yi-34b-chat.Q5_K_M.gguf"
   },
   "parameters": {
     "temperature": 0.7,
@@ -26,4 +32,3 @@
   },
   "engine": "nitro"
 }
-