chore: shared cuda dependencies

.husky/pre-commit

@@ -1 +1 @@
-oxlint --fix || npm run lint --fix
+npx oxlint@latest --fix

electron/package.json

@@ -18,7 +18,8 @@
       "docs/**/*",
       "scripts/**/*",
       "icons/**/*",
-      "themes"
+      "themes",
+      "shared"
     ],
     "asarUnpack": [
       "pre-install",
@@ -26,7 +27,8 @@
       "docs",
       "scripts",
       "icons",
-      "themes"
+      "themes",
+      "shared"
     ],
     "publish": [
       {

extensions/inference-cortex-extension/download.bat

@@ -1,5 +1,6 @@
 @echo off
 set BIN_PATH=./bin
+set SHARED_PATH=./../../electron/shared
 set /p CORTEX_VERSION=<./bin/version.txt
 
 @REM Download cortex.llamacpp binaries
@@ -18,8 +19,8 @@ call .\node_modules\.bin\download %DOWNLOAD_URL%-avx.tar.gz -e --strip 1 -o %BIN
 call .\node_modules\.bin\download %DOWNLOAD_URL%-avx2.tar.gz -e --strip 1 -o %BIN_PATH%/avx2/engines/cortex.llamacpp
 call .\node_modules\.bin\download %DOWNLOAD_URL%-avx512.tar.gz -e --strip 1 -o %BIN_PATH%/avx512/engines/cortex.llamacpp
 call .\node_modules\.bin\download %DOWNLOAD_URL%-vulkan.tar.gz -e --strip 1 -o %BIN_PATH%/vulkan/engines/cortex.llamacpp
-call .\node_modules\.bin\download %CUDA_DOWNLOAD_URL%/cuda-12-0-windows-amd64.tar.gz -e --strip 1 -o %BIN_PATH%
-call .\node_modules\.bin\download %CUDA_DOWNLOAD_URL%/cuda-11-7-windows-amd64.tar.gz -e --strip 1 -o %BIN_PATH%
+call .\node_modules\.bin\download %CUDA_DOWNLOAD_URL%/cuda-12-0-windows-amd64.tar.gz -e --strip 1 -o %SHARED_PATH%
+call .\node_modules\.bin\download %CUDA_DOWNLOAD_URL%/cuda-11-7-windows-amd64.tar.gz -e --strip 1 -o %SHARED_PATH%
 
 
 @REM Loop through each folder and move DLLs (excluding engine.dll)

extensions/inference-cortex-extension/download.sh

@@ -23,8 +23,8 @@ if [ "$OS_TYPE" == "Linux" ]; then
     download "${ENGINE_DOWNLOAD_URL}-linux-amd64-noavx-cuda-12-0.tar.gz" -e --strip 1 -o "./bin/noavx-cuda-12-0/engines/cortex.llamacpp" 1
     download "${ENGINE_DOWNLOAD_URL}-linux-amd64-noavx-cuda-11-7.tar.gz" -e --strip 1 -o "./bin/noavx-cuda-11-7/engines/cortex.llamacpp" 1
     download "${ENGINE_DOWNLOAD_URL}-linux-amd64-vulkan.tar.gz" -e --strip 1 -o "./bin/vulkan/engines/cortex.llamacpp" 1
-    download "${CUDA_DOWNLOAD_URL}/cuda-12-0-linux-amd64.tar.gz" -e --strip 1 -o "./bin" 1
-    download "${CUDA_DOWNLOAD_URL}/cuda-11-7-linux-amd64.tar.gz" -e --strip 1 -o "./bin" 1
+    download "${CUDA_DOWNLOAD_URL}/cuda-12-0-linux-amd64.tar.gz" -e --strip 1 -o "../../electron/shared" 1
+    download "${CUDA_DOWNLOAD_URL}/cuda-11-7-linux-amd64.tar.gz" -e --strip 1 -o "../../electron/shared" 1
 
 elif [ "$OS_TYPE" == "Darwin" ]; then
     # macOS downloads

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

@@ -2,6 +2,7 @@ import path from 'path'
 import { getJanDataFolderPath, log, SystemInformation } from '@janhq/core/node'
 import { executableCortexFile } from './execute'
 import { ProcessWatchdog } from './watchdog'
+import { appResourcePath } from '@janhq/core/node'
 
 // The HOST address to use for the Nitro subprocess
 const LOCAL_PORT = '39291'
@@ -19,9 +20,9 @@ function run(systemInfo?: SystemInformation): Promise<any> {
       // If ngl is not set or equal to 0, run on CPU with correct instructions
       systemInfo?.gpuSetting
         ? {
-            ...systemInfo.gpuSetting,
-            run_mode: systemInfo.gpuSetting.run_mode,
-          }
+          ...systemInfo.gpuSetting,
+          run_mode: systemInfo.gpuSetting.run_mode,
+        }
         : undefined
     )
 
@@ -29,20 +30,9 @@ function run(systemInfo?: SystemInformation): Promise<any> {
     log(`[CORTEX]:: Spawn cortex at path: ${executableOptions.executablePath}`)
     log(`[CORTEX]:: Cortex engine path: ${executableOptions.enginePath}`)
 
-    // Add engine path to the PATH and LD_LIBRARY_PATH
-    process.env.PATH = (process.env.PATH || '').concat(
-      path.delimiter,
-      executableOptions.enginePath,
-      path.delimiter,
-      executableOptions.binPath
-    )
-    log(`[CORTEX] PATH: ${process.env.PATH}`)
-    process.env.LD_LIBRARY_PATH = (process.env.LD_LIBRARY_PATH || '').concat(
-      path.delimiter,
-      executableOptions.enginePath,
-      path.delimiter,
-      executableOptions.binPath
-    )
+    addEnvPaths(path.join(appResourcePath(), 'shared'))
+    addEnvPaths(executableOptions.binPath)
+    addEnvPaths(executableOptions.enginePath)
 
     const dataFolderPath = getJanDataFolderPath()
     watchdog = new ProcessWatchdog(
@@ -83,6 +73,23 @@ function dispose() {
   watchdog?.terminate()
 }
 
+function addEnvPaths(dest: string) {
+  // Add engine path to the PATH and LD_LIBRARY_PATH
+  if (process.platform === 'win32') {
+    process.env.PATH = (process.env.PATH || '').concat(
+      path.delimiter,
+      dest,
+    )
+    log(`[CORTEX] PATH: ${process.env.PATH}`)
+  } else {
+    process.env.LD_LIBRARY_PATH = (process.env.LD_LIBRARY_PATH || '').concat(
+      path.delimiter,
+      dest,
+    )
+    log(`[CORTEX] LD_LIBRARY_PATH: ${process.env.LD_LIBRARY_PATH}`)
+  }
+}
+
 /**
  * Cortex process info
  */

extensions/model-extension/src/index.ts

@@ -13,6 +13,7 @@ import {
 import { CortexAPI } from './cortex'
 import { scanModelsFolder } from './legacy/model-json'
 import { downloadModel } from './legacy/download'
+import { systemInformation } from '@janhq/core'
 
 declare const SETTINGS: Array<any>
 
@@ -65,7 +66,7 @@ export default class JanModelExtension extends ModelExtension {
         model.engine === InferenceEngine.nitro_tensorrt_llm ||
         model.settings.vision_model
       ) {
-        return downloadModel(model)
+        return downloadModel(model, (await systemInformation()).gpuSetting)
       }
     }
     /**