foundry model run deepseek-r1-14b

[jasonwtli]

Ran it for a few models including phi & deepseek-r1-7b. But, doesn't work for the deepseek-r1-14b model (running on new Copilot+PCs with Snapdragon X-Elite chips).

Have tried all the usual troubleshoot. For Example: Service Start, reboot, clear cache and re-download. So, think its specific to 14b parameter model.

🕚 Loading model...
Exception: Request to local service failed. Uri:http://localhost:5273/openai/load/deepseek-r1-distill-qwen-14b-qnn-npu?ttl=600
An error occurred while sending the request.
Please check service status with 'foundry service status'.

[samuel100]

Thanks for raising an issue @jasonwtli - would you be able to upload the log files here:

foundry service diag —logs

This will create a zip file on your desktop to upload.

Note

Log files may contain information like user names, IP addresses, file paths, etc. Be sure to remove those before sharing here.

[jasonwtli]

Please see attached @samuel100 Thanks!

foundry20250601.log

[samuel100]

Thanks @jasonwtli - I think your NPU is running out memory. Does your NPU have ~8GB? This would explain why the 7B model (~3.7GB) works but not the 14B model (~7.2GB).

[jasonwtli]

Thanks @samuel100 Do you know why on the new Copilot+ Laptops with 16GB on Elite X chips have standard 7.8GB unified memory allocated to NPU? I have both an Asus & Lenovo and it should work as its optimized for these workloads? But, it doesn't and gave up tweaking the settings on the Asus as it just max out or throttled close to it.

However, tested it on my Windows 11 Qualcomm Snapdragon SDK and it works as it has the full 16GB of unified memory.

[leestott]

@samuel100 also confirmed this is an error on Asus Vivobook ARM Device

Issue with Model Deepseek R1 14b

Model Load Error

As you can see you dont really get any error indicating memory error? In the Diag

C:\Users\DPEGA>foundry service diag
===================================================
[System] %USERPROFILE%: C:\Users\DPEGA
[System] Environment.ProcessPath: C:\Program Files\WindowsApps\Microsoft.FoundryLocal_0.3.9267.43123_arm64__8wekyb3d8bbwe\foundry.exe
[System] Environment.ProcessId: 18268
[System] Environment.Is64BitProcess: True
[System] Environment.Is64BitOperatingSystem: True
[System] Environment.IsPrivilegedProcess: False
[System] Environment.CurrentDirectory: C:\Users\DPEGA
[System] Environment.SystemDirectory: C:\WINDOWS\system32
[System] OperatingSystem.IsWindows: True
[System] OperatingSystem.IsLinux: False
[System] OperatingSystem.IsMacOS: False
[System] Path.GetFullPath(.): C:\Users\DPEGA
[System] IPGlobalProperties.HostName: qualcomm
[System] IPGlobalProperties.DomainName:
[System] TcpListeners.Length: 26
===================================================
[FL CLI] Folder: C:\Program Files\WindowsApps\Microsoft.FoundryLocal_0.3.9267.43123_arm64__8wekyb3d8bbwe
[FL CLI] Config:
{
  "defaultLogLevel": 2,
  "serviceSettings": {
    "host": "localhost",
    "port": 5273,
    "cacheDirectoryPath": "C:\\Users\\DPEGA\\.foundry\\cache\\models",
    "schema": "http",
    "pipeName": "inference_agent",
    "defaultSecondsForModelTTL": 600
  }
}
===================================================