Executorh Qnn backend fastrpc error

[Jiawe-iLiu]

🐛 Describe the bug

I was following this tutorial to deploy the llama3 8b module on my 16G ram 8gen3 android phone. But I got the fastrpc issue when I was running the runner on the phone.

Versions

The log was:

[WARNING] [Qnn ExecuTorch]: This option (2) is only for offline prepare case.

[INFO] [Qnn ExecuTorch]: Running level=3 optimization.
[INFO] [Qnn ExecuTorch]: create QNN Logger with log_level 2
[INFO] [Qnn ExecuTorch]: Initialize Qnn backend parameters for Qnn executorch backend type 2
[INFO] [Qnn ExecuTorch]: Caching: Caching is in RESTORE MODE.
[WARNING] [Qnn ExecuTorch]: Function not called, PrepareLib isn't loaded!

[ERROR] [Qnn ExecuTorch]: fastrpc memory map for fd: 58 with length: 440401920 failed with error: 0x1

[ERROR] [Qnn ExecuTorch]: Failed to map weights buffer to device!

[ERROR] [Qnn ExecuTorch]: Could not allocate persistent weights buffer!

[ERROR] [Qnn ExecuTorch]: Failed to initialize graph memory

[ERROR] [Qnn ExecuTorch]: Failed to initialize graph with id 258 context 7 deviceId 0 coreId 0 pdId 0 with err 1002

[ERROR] [Qnn ExecuTorch]: Context create from binary failed for deviceId 0 coreId 0 pdId 0 for context 7, err 1002

[ERROR] [Qnn ExecuTorch]: Trying context create from binary on new PD for contextId 7

[WARNING] [Qnn ExecuTorch]: Function not called, PrepareLib isn't loaded!

[WARNING] [Qnn ExecuTorch]: This option (2) is only for offline prepare case.

Does anyone have the same issue before?

cc @cccclai

[shewu-quic]

Hi @Jiawe-iLiu ,
Thanks for your trying llama3 8B with QNN Backend.

Based on your log, there is no problem. Maybe [ERROR] makes you a little confused.

[ERROR] [Qnn ExecuTorch]: Trying context create from binary on new PD for contextId 7

This message indicates that a new PD is being created to load the model.

Maybe you could provide more log to clarify the issue.