Merge branch 'dev' into patch-1

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

@@ -114,15 +114,14 @@ jobs:
       - name: Upload latest-mac.yml
         if: ${{ needs.set-public-provider.outputs.public_provider == 'aws-s3' }}
         run: |
-          aws s3 cp ./latest-mac.yml "s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/temp-latest/latest-mac.yml"
+          aws s3 cp ./latest-mac.yml "s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/temp-nightly/latest-mac.yml"
-          aws s3 sync s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/temp-latest/ s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/latest/
+          aws s3 sync s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/temp-nightly/ s3://${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}/nightly/
         env:
           AWS_ACCESS_KEY_ID: ${{ secrets.DELTA_AWS_ACCESS_KEY_ID }}
           AWS_SECRET_ACCESS_KEY: ${{ secrets.DELTA_AWS_SECRET_ACCESS_KEY }}
           AWS_DEFAULT_REGION: ${{ secrets.DELTA_AWS_REGION }}
           AWS_EC2_METADATA_DISABLED: "true"
 
-    
   noti-discord-nightly-and-update-url-readme:
     needs: [build-macos-x64, build-macos-arm64, build-windows-x64, build-linux-x64, get-update-version, set-public-provider, combine-latest-mac-yml]
     secrets: inherit

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

@@ -60,16 +60,16 @@ jobs:
           mv /tmp/package.json electron/package.json
           jq --arg version "${{ inputs.new_version }}" '.version = $version' web/package.json > /tmp/package.json
           mv /tmp/package.json web/package.json
-          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/latest", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-latest", "channel": "latest"}]' electron/package.json > /tmp/package.json
+          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/nightly", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-nightly", "channel": "latest"}]' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
           cat electron/package.json
-          # chmod +x .github/scripts/rename-app.sh
+          chmod +x .github/scripts/rename-app.sh
-          # .github/scripts/rename-app.sh ./electron/package.json nightly
+          .github/scripts/rename-app.sh ./electron/package.json nightly
-          # chmod +x .github/scripts/rename-workspace.sh
+          chmod +x .github/scripts/rename-workspace.sh
-          # .github/scripts/rename-workspace.sh ./package.json nightly
+          .github/scripts/rename-workspace.sh ./package.json nightly
-          # echo "------------------------"
+          echo "------------------------"
-          # cat ./electron/package.json
+          cat ./electron/package.json
-          # echo "------------------------"
+          echo "------------------------"
 
       - name: Change App Name for beta version
         if: inputs.beta == true

.github/workflows/template-build-macos-arm64.yml

@@ -72,20 +72,20 @@ jobs:
           jq --arg version "${{ inputs.new_version }}" '.version = $version' web/package.json > /tmp/package.json
           mv /tmp/package.json web/package.json
 
-          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/latest", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-latest", "channel": "latest"}]' electron/package.json > /tmp/package.json
+          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/nightly", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-nightly", "channel": "latest"}]' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
 
           jq --arg teamid "${{ secrets.APPLE_TEAM_ID }}" '.build.mac.notarize.teamId = $teamid' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
 
           # cat electron/package.json
-          # chmod +x .github/scripts/rename-app.sh
+          chmod +x .github/scripts/rename-app.sh
-          # .github/scripts/rename-app.sh ./electron/package.json nightly
+          .github/scripts/rename-app.sh ./electron/package.json nightly
-          # chmod +x .github/scripts/rename-workspace.sh
+          chmod +x .github/scripts/rename-workspace.sh
-          # .github/scripts/rename-workspace.sh ./package.json nightly
+          .github/scripts/rename-workspace.sh ./package.json nightly
-          # echo "------------------------"
+          echo "------------------------"
-          # cat ./electron/package.json
+          cat ./electron/package.json
-          # echo "------------------------"
+          echo "------------------------"
 
       - name: Change App Name for beta version
         if: inputs.beta == true

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

@@ -72,20 +72,20 @@ jobs:
           jq --arg version "${{ inputs.new_version }}" '.version = $version' web/package.json > /tmp/package.json
           mv /tmp/package.json web/package.json
 
-          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/latest", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-latest", "channel": "latest"}]' electron/package.json > /tmp/package.json
+          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/nightly", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-nightly", "channel": "latest"}]' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
 
           jq --arg teamid "${{ secrets.APPLE_TEAM_ID }}" '.build.mac.notarize.teamId = $teamid' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
 
-          # cat electron/package.json
+          cat electron/package.json
-          # chmod +x .github/scripts/rename-app.sh
+          chmod +x .github/scripts/rename-app.sh
-          # .github/scripts/rename-app.sh ./electron/package.json nightly
+          .github/scripts/rename-app.sh ./electron/package.json nightly
-          # chmod +x .github/scripts/rename-workspace.sh
+          chmod +x .github/scripts/rename-workspace.sh
-          # .github/scripts/rename-workspace.sh ./package.json nightly
+          .github/scripts/rename-workspace.sh ./package.json nightly
-          # echo "------------------------"
+          echo "------------------------"
-          # cat ./electron/package.json
+          cat ./electron/package.json
-          # echo "------------------------"
+          echo "------------------------"
 
       - name: Change App Name for beta version
         if: inputs.beta == true

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

@@ -73,24 +73,24 @@ jobs:
           jq --arg version "${{ inputs.new_version }}" '.version = $version' web/package.json > /tmp/package.json
           mv /tmp/package.json web/package.json
 
-          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/latest", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-latest", "channel": "latest"}]' electron/package.json > /tmp/package.json
+          jq '.build.publish = [{"provider": "generic", "url": "https://delta.jan.ai/nightly", "channel": "latest"}, {"provider": "s3", "acl": null, "bucket": "${{ secrets.DELTA_AWS_S3_BUCKET_NAME }}", "region": "${{ secrets.DELTA_AWS_REGION}}", "path": "temp-nightly", "channel": "latest"}]' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
 
           jq '.build.win.sign = "./sign.js"' electron/package.json > /tmp/package.json
           mv /tmp/package.json electron/package.json
           cat electron/package.json
 
-          # chmod +x .github/scripts/rename-app.sh
+          chmod +x .github/scripts/rename-app.sh
-          # .github/scripts/rename-app.sh ./electron/package.json nightly
+          .github/scripts/rename-app.sh ./electron/package.json nightly
-          # chmod +x .github/scripts/rename-workspace.sh
+          chmod +x .github/scripts/rename-workspace.sh
-          # .github/scripts/rename-workspace.sh ./package.json nightly
+          .github/scripts/rename-workspace.sh ./package.json nightly
-          # chmod +x .github/scripts/rename-uninstaller.sh
+          chmod +x .github/scripts/rename-uninstaller.sh
-          # .github/scripts/rename-uninstaller.sh nightly
+          .github/scripts/rename-uninstaller.sh nightly
-          # echo "------------------------"
+          echo "------------------------"
-          # cat ./electron/package.json
+          cat ./electron/package.json
-          # echo "------------------------"
+          echo "------------------------"
-          # cat ./package.json
+          cat ./package.json
-          # echo "------------------------"
+          echo "------------------------"
 
       - name: Change App Name for beta version
         if: inputs.beta == true

.github/workflows/template-noti-discord-and-update-url-readme.yml

@@ -47,11 +47,11 @@ jobs:
         with:
           args: |
             Jan App ${{ inputs.build_reason }} build artifact version {{ VERSION }}:
-            - Windows: https://delta.jan.ai/latest/jan-win-x64-{{ VERSION }}.exe
+            - Windows: https://delta.jan.ai/nightly/jan-win-x64-{{ VERSION }}.exe
-            - macOS Intel: https://delta.jan.ai/latest/jan-mac-x64-{{ VERSION }}.dmg
+            - macOS Intel: https://delta.jan.ai/nightly/jan-mac-x64-{{ VERSION }}.dmg
-            - macOS Apple Silicon: https://delta.jan.ai/latest/jan-mac-arm64-{{ VERSION }}.dmg
+            - macOS Apple Silicon: https://delta.jan.ai/nightly/jan-mac-arm64-{{ VERSION }}.dmg
-            - Linux Deb: https://delta.jan.ai/latest/jan-linux-amd64-{{ VERSION }}.deb
+            - Linux Deb: https://delta.jan.ai/nightly/jan-linux-amd64-{{ VERSION }}.deb
-            - Linux AppImage: https://delta.jan.ai/latest/jan-linux-x86_64-{{ VERSION }}.AppImage
+            - Linux AppImage: https://delta.jan.ai/nightly/jan-linux-x86_64-{{ VERSION }}.AppImage
             - Github action run: https://github.com/janhq/jan/actions/runs/{{ GITHUB_RUN_ID }}
         env:
           DISCORD_WEBHOOK: ${{ secrets.DISCORD_WEBHOOK }}

core/src/node/api/processors/download.ts

@@ -1,6 +1,6 @@
 import { resolve, sep } from 'path'
 import { DownloadEvent } from '../../../types/api'
-import { normalizeFilePath, validatePath } from '../../helper/path'
+import { normalizeFilePath } from '../../helper/path'
 import { getJanDataFolderPath } from '../../helper'
 import { DownloadManager } from '../../helper/download'
 import { createWriteStream, renameSync } from 'fs'
@@ -37,7 +37,6 @@ export class Downloader implements Processor {
     const modelId = downloadRequest.modelId ?? array.pop() ?? ''
 
     const destination = resolve(getJanDataFolderPath(), normalizedPath)
-    validatePath(destination)
     const rq = request({ url, strictSSL, proxy })
 
     // Put request to download manager instance

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

@@ -1,5 +1,5 @@
 import { join, resolve } from 'path'
-import { normalizeFilePath, validatePath } from '../../helper/path'
+import { normalizeFilePath } from '../../helper/path'
 import { getJanDataFolderPath } from '../../helper'
 import { Processor } from './Processor'
 import fs from 'fs'
@@ -36,7 +36,6 @@ export class FileSystem implements Processor {
               return path
             }
             const absolutePath = resolve(path)
-            validatePath(absolutePath)
             return absolutePath
           })
         )
@@ -55,7 +54,6 @@ export class FileSystem implements Processor {
     }
 
     const absolutePath = resolve(path)
-    validatePath(absolutePath)
 
     return new Promise((resolve, reject) => {
       fs.rm(absolutePath, { recursive: true, force: true }, (err) => {
@@ -79,7 +77,6 @@ export class FileSystem implements Processor {
     }
 
     const absolutePath = resolve(path)
-    validatePath(absolutePath)
 
     return new Promise((resolve, reject) => {
       fs.mkdir(absolutePath, { recursive: true }, (err) => {

core/src/node/api/processors/fsExt.ts

@@ -1,6 +1,6 @@
 import { basename, join } from 'path'
 import fs, { readdirSync } from 'fs'
-import { appResourcePath, normalizeFilePath, validatePath } from '../../helper/path'
+import { appResourcePath, normalizeFilePath } from '../../helper/path'
 import { defaultAppConfig, getJanDataFolderPath, getJanDataFolderPath as getPath } from '../../helper'
 import { Processor } from './Processor'
 import { FileStat } from '../../../types'
@@ -61,7 +61,6 @@ export class FSExt implements Processor {
       
       const dataBuffer = Buffer.from(data, 'base64')
       const writePath = join(getJanDataFolderPath(), normalizedPath)
-      validatePath(writePath)
       fs.writeFileSync(writePath, dataBuffer)
     } catch (err) {
       console.error(`writeFile ${path} result: ${err}`)
@@ -69,7 +68,6 @@ export class FSExt implements Processor {
   }
 
   copyFile(src: string, dest: string): Promise<void> {
-    validatePath(dest)
     return new Promise((resolve, reject) => {
       fs.copyFile(src, dest, (err) => {
         if (err) {

core/src/node/helper/path.ts

@@ -34,18 +34,4 @@ export function appResourcePath() {
 
   // server
   return join(global.core.appPath(), '../../..')
-}
+}
-
-export function validatePath(path: string) {
-  const appDataFolderPath = getJanDataFolderPath()
-  const resourcePath = appResourcePath()
-  const applicationSupportPath = global.core?.appPath() ?? resourcePath
-  const absolutePath = resolve(__dirname, path)
-  if (
-    ![appDataFolderPath, resourcePath, applicationSupportPath].some((whiteListedPath) =>
-      absolutePath.startsWith(whiteListedPath)
-    )
-  ) {
-    throw new Error(`Invalid path: ${absolutePath}`)
-  }
-}

electron/utils/migration.ts

@@ -61,6 +61,9 @@ async function checkAndMigrateTheme(
   )
   if (existingTheme) {
     const desTheme = join(janDataThemesFolder, existingTheme)
+    if (!lstatSync(desTheme).isDirectory()) {
+      return
+    }
     console.debug('Updating theme', existingTheme)
     rmdirSync(desTheme, { recursive: true })
     cpSync(sourceThemePath, join(janDataThemesFolder, sourceThemeName), {

extensions/inference-cortex-extension/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@janhq/inference-cortex-extension",
   "productName": "Cortex Inference Engine",
-  "version": "1.0.20",
+  "version": "1.0.21",
   "description": "This extension embeds cortex.cpp, a lightweight inference engine written in C++. See https://jan.ai.\nAdditional dependencies could be installed to run without Cuda Toolkit installation.",
   "main": "dist/index.js",
   "node": "dist/node/index.cjs.js",

extensions/inference-cortex-extension/resources/models/qwen2.5-coder-14b-instruct/model.json

@@ -0,0 +1,36 @@
+{
+    "sources": [
+      {
+        "filename": "Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf",
+        "url": "https://huggingface.co/bartowski/Qwen2.5-Coder-14B-Instruct-GGUF/resolve/main/Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf"
+      }
+    ],
+    "id": "qwen2.5-coder-14b-instruct",
+    "object": "model",
+    "name": "Qwen2.5 Coder 14B Instruct Q4",
+    "version": "1.0",
+    "description": "Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models. Significantly improvements in code generation, code reasoning and code fixing.",
+    "format": "gguf",
+    "settings": {
+      "ctx_len": 32768,
+      "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": "Qwen2.5-Coder-14B-Instruct-Q4_K_M.gguf",
+      "ngl": 29
+    },
+    "parameters": {
+      "temperature": 0.7,
+      "top_p": 0.95,
+      "stream": true,
+      "max_tokens": 32768,
+      "stop": ["<|endoftext|>", "<|im_end|>"],
+      "frequency_penalty": 0,
+      "presence_penalty": 0
+    },
+    "metadata": {
+      "author": "QwenLM",
+      "tags": ["14B", "Featured"],
+      "size": 8990000000
+    },
+    "engine": "llama-cpp"
+  }
+  

extensions/inference-cortex-extension/resources/models/qwen2.5-coder-32b-instruct/model.json

@@ -0,0 +1,36 @@
+{
+    "sources": [
+      {
+        "filename": "Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf",
+        "url": "https://huggingface.co/bartowski/Qwen2.5-Coder-32B-Instruct-GGUF/resolve/main/Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf"
+      }
+    ],
+    "id": "qwen2.5-coder-32b-instruct",
+    "object": "model",
+    "name": "Qwen2.5 Coder 32B Instruct Q4",
+    "version": "1.0",
+    "description": "Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models. Significantly improvements in code generation, code reasoning and code fixing.",
+    "format": "gguf",
+    "settings": {
+      "ctx_len": 32768,
+      "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": "Qwen2.5-Coder-32B-Instruct-Q4_K_M.gguf",
+      "ngl": 29
+    },
+    "parameters": {
+      "temperature": 0.7,
+      "top_p": 0.95,
+      "stream": true,
+      "max_tokens": 32768,
+      "stop": ["<|endoftext|>", "<|im_end|>"],
+      "frequency_penalty": 0,
+      "presence_penalty": 0
+    },
+    "metadata": {
+      "author": "QwenLM",
+      "tags": ["32B", "Featured"],
+      "size": 19900000000
+    },
+    "engine": "llama-cpp"
+  }
+  

extensions/inference-cortex-extension/rollup.config.ts

@@ -49,6 +49,8 @@ const llama321bJson = require('./resources/models/llama3.2-1b-instruct/model.jso
 const llama323bJson = require('./resources/models/llama3.2-3b-instruct/model.json')
 const qwen257bJson = require('./resources/models/qwen2.5-7b-instruct/model.json')
 const qwen25coder7bJson = require('./resources/models/qwen2.5-coder-7b-instruct/model.json')
+const qwen25coder14bJson = require('./resources/models/qwen2.5-coder-14b-instruct/model.json')
+const qwen25coder32bJson = require('./resources/models/qwen2.5-coder-32b-instruct/model.json')
 const qwen2514bJson = require('./resources/models/qwen2.5-14b-instruct/model.json')
 const qwen2532bJson = require('./resources/models/qwen2.5-32b-instruct/model.json')
 const qwen2572bJson = require('./resources/models/qwen2.5-72b-instruct/model.json')
@@ -108,6 +110,8 @@ export default [
           llama323bJson,
           qwen257bJson,
           qwen25coder7bJson,
+          qwen25coder14bJson,
+          qwen25coder32bJson,
           qwen2514bJson,
           qwen2532bJson,
           qwen2572bJson,
@@ -115,6 +119,7 @@ export default [
         NODE: JSON.stringify(`${packageJson.name}/${packageJson.node}`),
         DEFAULT_SETTINGS: JSON.stringify(defaultSettingJson),
         CORTEX_API_URL: JSON.stringify('http://127.0.0.1:39291'),
+        CORTEX_SOCKET_URL: JSON.stringify('ws://127.0.0.1:39291'),
       }),
       // Allow json resolution
       json(),

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

@@ -1,5 +1,6 @@
 declare const NODE: string
 declare const CORTEX_API_URL: string
+declare const CORTEX_SOCKET_URL: string
 declare const DEFAULT_SETTINGS: Array<any>
 declare const MODELS: Array<any>
 

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

@@ -16,17 +16,29 @@ import {
   getJanDataFolderPath,
   extractModelLoadParams,
   fs,
+  events,
+  ModelEvent
 } from '@janhq/core'
 import PQueue from 'p-queue'
 import ky from 'ky'
 
+/**
+ * Event subscription types of Downloader
+ */
+enum DownloadTypes {
+  DownloadUpdated = 'onFileDownloadUpdate',
+  DownloadError = 'onFileDownloadError',
+  DownloadSuccess = 'onFileDownloadSuccess',
+  DownloadStopped = 'onFileDownloadStopped',
+  DownloadStarted = 'onFileDownloadStarted',
+}
+
 /**
  * A class that implements the InferenceExtension interface from the @janhq/core package.
  * The class provides methods for initializing and stopping a model, and for making inference requests.
  * It also subscribes to events emitted by the @janhq/core package and handles new message requests.
  */
 export default class JanInferenceCortexExtension extends LocalOAIEngine {
-  // DEPRECATED
   nodeModule: string = 'node'
 
   queue = new PQueue({ concurrency: 1 })
@@ -38,6 +50,11 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
    */
   inferenceUrl = `${CORTEX_API_URL}/v1/chat/completions`
 
+  /**
+   * Socket instance of events subscription
+   */
+  socket?: WebSocket = undefined
+
   /**
    * Subscribes to events emitted by the @janhq/core package.
    */
@@ -55,6 +72,8 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
 
     this.queue.add(() => this.healthz())
 
+    this.subscribeToEvents()
+
     window.addEventListener('beforeunload', () => {
       this.clean()
     })
@@ -138,7 +157,7 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
           methods: ['get'],
         },
       })
-      .then(() => {})
+      .then(() => { })
   }
 
   /**
@@ -154,6 +173,50 @@ export default class JanInferenceCortexExtension extends LocalOAIEngine {
         // Do nothing
       })
   }
+
+  /**
+   * Subscribe to cortex.cpp websocket events
+   */
+  subscribeToEvents() {
+    this.queue.add(
+      () =>
+        new Promise<void>((resolve) => {
+          this.socket = new WebSocket(`${CORTEX_SOCKET_URL}/events`)
+
+          this.socket.addEventListener('message', (event) => {
+            const data = JSON.parse(event.data)
+            const transferred = data.task.items.reduce(
+              (acc: number, cur: any) => acc + cur.downloadedBytes,
+              0
+            )
+            const total = data.task.items.reduce(
+              (acc: number, cur: any) => acc + cur.bytes,
+              0
+            )
+            const percent = total > 0 ? transferred / total : 0
+
+            events.emit(DownloadTypes[data.type as keyof typeof DownloadTypes], {
+              modelId: data.task.id,
+              percent: percent,
+              size: {
+                transferred: transferred,
+                total: total,
+              },
+            })
+            // Update models list from Hub
+            if (data.type === DownloadTypes.DownloadSuccess) {
+              // Delay for the state update from cortex.cpp
+              // Just to be sure
+              setTimeout(() => {
+                events.emit(ModelEvent.OnModelsUpdate, {})
+              }, 500)
+            }
+          })
+          resolve()
+        })
+    )
+  }
+
 }
 
 /// Legacy

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

@@ -70,16 +70,17 @@ export default class JanInferenceOpenAIExtension extends RemoteOAIEngine {
    * Tranform the payload before sending it to the inference endpoint.
    * The new preview models such as o1-mini and o1-preview replaced max_tokens by max_completion_tokens parameter.
    * Others do not.
-   * @param payload 
+   * @param payload
-   * @returns 
+   * @returns
    */
   transformPayload = (payload: OpenAIPayloadType): OpenAIPayloadType => {
     // Transform the payload for preview models
     if (this.previewModels.includes(payload.model)) {
-      const { max_tokens, ...params } = payload
+      const { max_tokens, temperature, top_p, stop, ...params } = payload
       return {
         ...params,
         max_completion_tokens: max_tokens,
+        stream: false // o1 only support stream = false
       }
     }
     // Pass through for non-preview models

extensions/model-extension/src/cortex.ts

@@ -1,6 +1,6 @@
 import PQueue from 'p-queue'
 import ky from 'ky'
-import { events, extractModelLoadParams, Model, ModelEvent } from '@janhq/core'
+import {  extractModelLoadParams, Model } from '@janhq/core'
 import { extractInferenceParams } from '@janhq/core'
 /**
  * cortex.cpp Model APIs interface
@@ -24,21 +24,11 @@ type ModelList = {
   data: any[]
 }
 
-enum DownloadTypes {
-  DownloadUpdated = 'onFileDownloadUpdate',
-  DownloadError = 'onFileDownloadError',
-  DownloadSuccess = 'onFileDownloadSuccess',
-  DownloadStopped = 'onFileDownloadStopped',
-  DownloadStarted = 'onFileDownloadStarted',
-}
-
 export class CortexAPI implements ICortexAPI {
   queue = new PQueue({ concurrency: 1 })
-  socket?: WebSocket = undefined
 
   constructor() {
     this.queue.add(() => this.healthz())
-    this.subscribeToEvents()
   }
 
   /**
@@ -172,49 +162,6 @@ export class CortexAPI implements ICortexAPI {
       .then(() => {})
   }
 
-  /**
-   * Subscribe to cortex.cpp websocket events
-   */
-  subscribeToEvents() {
-    this.queue.add(
-      () =>
-        new Promise<void>((resolve) => {
-          this.socket = new WebSocket(`${SOCKET_URL}/events`)
-
-          this.socket.addEventListener('message', (event) => {
-            const data = JSON.parse(event.data)
-            const transferred = data.task.items.reduce(
-              (acc, cur) => acc + cur.downloadedBytes,
-              0
-            )
-            const total = data.task.items.reduce(
-              (acc, cur) => acc + cur.bytes,
-              0
-            )
-            const percent = total > 0 ? transferred / total : 0
-
-            events.emit(DownloadTypes[data.type], {
-              modelId: data.task.id,
-              percent: percent,
-              size: {
-                transferred: transferred,
-                total: total,
-              },
-            })
-            // Update models list from Hub
-            if (data.type === DownloadTypes.DownloadSuccess) {
-              // Delay for the state update from cortex.cpp
-              // Just to be sure
-              setTimeout(() => {
-                events.emit(ModelEvent.OnModelsUpdate, {})
-              }, 500)
-            }
-          })
-          resolve()
-        })
-    )
-  }
-
   /**
    * TRansform model to the expected format (e.g. parameters, settings, metadata)
    * @param model

extensions/model-extension/src/index.ts

@@ -179,8 +179,8 @@ export default class JanModelExtension extends ModelExtension {
     if (toImportModels.length > 0) {
       // Import models
       await Promise.all(
-        toImportModels.map(async (model: Model & { file_path: string }) =>
+        toImportModels.map(async (model: Model & { file_path: string }) => {
-          this.importModel(
+          return this.importModel(
             model.id,
             model.sources[0].url.startsWith('http') ||
               !(await fs.existsSync(model.sources[0].url))
@@ -200,7 +200,7 @@ export default class JanModelExtension extends ModelExtension {
               ...model.parameters,
             } as Partial<Model>)
           })
-        )
+        })
       )
 
       return currentModels

web/containers/Layout/BottomPanel/SystemMonitor/index.tsx

@@ -79,7 +79,7 @@ const SystemMonitor = () => {
       {showSystemMonitorPanel && (
         <div
           className={twMerge(
-            'fixed bottom-9 left-[49px] z-50 flex h-[200px] w-[calc(100%-48px)] flex-shrink-0 flex-col border-t border-[hsla(var(--app-border))] bg-[hsla(var(--app-bg))]',
+            'system-monitor-panel fixed bottom-9 left-[49px] z-50 flex w-[calc(100%-48px)] flex-shrink-0 flex-col border-t border-[hsla(var(--app-border))] bg-[hsla(var(--app-bg))]',
             showFullScreen && 'h-[calc(100%-63px)]',
             reduceTransparent && 'w-[calc(100%-48px)] rounded-none'
           )}
@@ -147,7 +147,6 @@ const SystemMonitor = () => {
                   <span className="flex-shrink-0 ">{ramUtilitized}%</span>
                 </div>
               </div>
-
               {gpus.length > 0 && (
                 <div className="mb-4 border-b border-[hsla(var(--app-border))] pb-4 last:border-none">
                   {gpus.map((gpu, index) => {

web/containers/ModelDropdown/index.tsx

@@ -108,6 +108,11 @@ const ModelDropdown = ({
   const filteredDownloadedModels = useMemo(
     () =>
       configuredModels
+        .concat(
+          downloadedModels.filter(
+            (e) => !configuredModels.some((x) => x.id === e.id)
+          )
+        )
         .filter((e) =>
           e.name.toLowerCase().includes(searchText.toLowerCase().trim())
         )

web/hooks/useDropModelBinaries.ts

@@ -29,7 +29,7 @@ export default function useDropModelBinaries() {
       const importingModels: ImportingModel[] = supportedFiles.map((file) => ({
         importId: uuidv4(),
         modelId: undefined,
-        name: file.name.replace('.gguf', ''),
+        name: file.name.replace(/ /g, '').replace('.gguf', ''),
         description: '',
         path: file.path,
         tags: [],

web/hooks/useImportModel.ts

@@ -5,14 +5,14 @@ import {
   ImportingModel,
   LocalImportModelEvent,
   Model,
-  ModelEvent,
   ModelExtension,
   OptionType,
   events,
   fs,
+  baseName,
 } from '@janhq/core'
 
-import { atom, useSetAtom } from 'jotai'
+import { atom, useAtomValue, useSetAtom } from 'jotai'
 
 import { v4 as uuidv4 } from 'uuid'
 
@@ -23,6 +23,7 @@ import { FilePathWithSize } from '@/utils/file'
 import { extensionManager } from '@/extension'
 import {
   addDownloadingModelAtom,
+  downloadedModelsAtom,
   importingModelsAtom,
   removeDownloadingModelAtom,
 } from '@/helpers/atoms/Model.atom'
@@ -58,11 +59,24 @@ const useImportModel = () => {
   const setImportingModels = useSetAtom(importingModelsAtom)
   const addDownloadingModel = useSetAtom(addDownloadingModelAtom)
   const removeDownloadingModel = useSetAtom(removeDownloadingModelAtom)
+  const downloadedModels = useAtomValue(downloadedModelsAtom)
+
+  const incrementalModelName = useCallback(
+    (name: string, startIndex: number = 0): string => {
+      const newModelName = startIndex ? `${name}-${startIndex}` : name
+      if (downloadedModels.some((model) => model.id === newModelName)) {
+        return incrementalModelName(name, startIndex + 1)
+      } else {
+        return newModelName
+      }
+    },
+    [downloadedModels]
+  )
 
   const importModels = useCallback(
     (models: ImportingModel[], optionType: OptionType) => {
-      models.map((model) => {
+      models.map(async (model) => {
-        const modelId = model.modelId ?? model.path.split('/').pop()
+        const modelId = model.modelId ?? incrementalModelName(model.name)
         if (modelId) {
           addDownloadingModel(modelId)
           extensionManager
@@ -78,7 +92,7 @@ const useImportModel = () => {
         }
       })
     },
-    [addDownloadingModel, removeDownloadingModel]
+    [addDownloadingModel, incrementalModelName, removeDownloadingModel]
   )
 
   const updateModelInfo = useCallback(
@@ -100,7 +114,7 @@ const useImportModel = () => {
         ({ path, name, size }: FilePathWithSize) => ({
           importId: uuidv4(),
           modelId: undefined,
-          name: name.replace('.gguf', ''),
+          name: name.replace(/ /g, '').replace('.gguf', ''),
           description: '',
           path: path,
           tags: [],

web/package.json

@@ -38,7 +38,7 @@
     "react": "18.2.0",
     "react-circular-progressbar": "^2.1.0",
     "react-dom": "18.2.0",
-    "react-dropzone": "^14.2.3",
+    "react-dropzone": "14.2.3",
     "react-hook-form": "^7.47.0",
     "react-hot-toast": "^2.4.1",
     "react-icons": "^4.12.0",

web/screens/Settings/Advanced/index.tsx

@@ -189,7 +189,7 @@ const Advanced = () => {
    * @param gpuId
    * @returns
    */
-  const handleGPUChange = (gpuId: string) => {
+  const handleGPUChange = async (gpuId: string) => {
     let updatedGpusInUse = [...gpusInUse]
     if (updatedGpusInUse.includes(gpuId)) {
       updatedGpusInUse = updatedGpusInUse.filter((id) => id !== gpuId)
@@ -208,7 +208,7 @@ const Advanced = () => {
       updatedGpusInUse.push(gpuId)
     }
     setGpusInUse(updatedGpusInUse)
-    saveSettings({ gpusInUse: updatedGpusInUse })
+    await saveSettings({ gpusInUse: updatedGpusInUse })
     window.core?.api?.relaunch()
   }
 
@@ -306,7 +306,13 @@ const Advanced = () => {
                           })
                         }
                         // Stop any running model to apply the changes
-                        if (e.target.checked !== gpuEnabled) stopModel()
+                        if (e.target.checked !== gpuEnabled) {
+                          stopModel().finally(() => {
+                            setTimeout(() => {
+                              window.location.reload()
+                            }, 300)
+                          })
+                        }
                       }}
                     />
                   }

web/screens/Settings/SelectingModelModal/index.tsx

@@ -21,7 +21,7 @@ const SelectingModelModal = () => {
 
   const onSelectFileClick = useCallback(async () => {
     const platform = (await systemInformation()).osInfo?.platform
-    if (platform === 'win32') {
+    if (platform !== 'darwin') {
       setImportModelStage('CHOOSE_WHAT_TO_IMPORT')
       return
     }

web/screens/Thread/ThreadCenterPanel/LoadModelError/index.tsx

@@ -64,49 +64,6 @@ const LoadModelError = () => {
           to continue using it.
         </p>
       )
-    } else if (
-      settings &&
-      settings.run_mode === 'gpu' &&
-      !settings.vulkan &&
-      (!settings.nvidia_driver?.exist || !settings.cuda?.exist)
-    ) {
-      return (
-        <>
-          {!settings?.cuda.exist ? (
-            <p>
-              The CUDA toolkit may be unavailable. Please use the{' '}
-              <span
-                className="cursor-pointer font-medium text-[hsla(var(--app-link))]"
-                onClick={() => {
-                  setMainState(MainViewState.Settings)
-                  if (activeThread?.assistants[0]?.model.engine) {
-                    const engine = EngineManager.instance().get(
-                      activeThread.assistants[0].model.engine
-                    )
-                    engine?.name && setSelectedSettingScreen(engine.name)
-                  }
-                }}
-              >
-                Install Additional Dependencies
-              </span>{' '}
-              setting to proceed with the download / installation process.
-            </p>
-          ) : (
-            <div>
-              Problem with Nvidia drivers. Please follow the{' '}
-              <a
-                className="font-medium text-[hsla(var(--app-link))]"
-                href="https://www.nvidia.com/Download/index.aspx"
-                target="_blank"
-              >
-                Nvidia Drivers guideline
-              </a>{' '}
-              to access installation instructions and ensure proper functioning
-              of the application.
-            </div>
-          )}
-        </>
-      )
     } else {
       return (
         <div>

web/screens/Thread/ThreadCenterPanel/index.tsx

@@ -147,6 +147,20 @@ const ThreadCenterPanel = () => {
 
   const showSystemMonitorPanel = useAtomValue(showSystemMonitorPanelAtom)
 
+  const [height, setHeight] = useState<number>(0)
+
+  useEffect(() => {
+    if (showSystemMonitorPanel) {
+      const element = document.querySelector('.system-monitor-panel')
+
+      if (element) {
+        setHeight(element.clientHeight) // You can also use offsetHeight if needed
+      }
+    } else {
+      setHeight(0)
+    }
+  }, [showSystemMonitorPanel])
+
   return (
     <CenterPanelContainer>
       <div
@@ -193,9 +207,10 @@ const ThreadCenterPanel = () => {
         )}
         <div
           className={twMerge(
-            'flex h-full w-full flex-col justify-between',
+            'flex h-full w-full flex-col justify-between'
-            showSystemMonitorPanel && 'h-[calc(100%-200px)]'
+            // showSystemMonitorPanel && `h-[calc(100%-${height}px)]`
           )}
+          style={{ height: `calc(100% - ${height}px)` }}
         >
           {activeThread ? (
             <div className="flex h-full w-full overflow-x-hidden">