I am using qnn2.10.0 to run the fp16 tutorial on the 8295 platform. I have referred to two documents, but I am still confused about how to execute the fp16 process. I made the following attempts and got error logs. Please help me troubleshoot the low-level errors. Thank you very much.
The qnn 2.10.0 version SDK failed to execute the fp16 tutorial on the 8295 platform with the HTP backend.
Posted: Thu, 2023-06-15 03:03
I am trying to run a command on Android using the QNN SDK. The command I am using is:
I received the following error log:
The command I am trying to run is based on the document located at:
The cache file I received is 46M,and 92M x86_64-linux-clang/libInception_v3.so.
The JSON config file I am using is:
Hi, did you solve this problem? I also met this issue.
E QNN: Skel failed to process context binary.
E QNN: Context create from binary failed with err 5005
E QNN: Could not create context from binary and this:
E QNN: Create From Binary failure
-------------------------------------------
from above logs, the environment is good, the issue come from v68Skel and binary. v68Skel will check whether binary match the hardware.
The config file you used seems fine also, one suspection is that you use mobile QNN SDK. fp16 for v68 is only enable in Auto QNN SDK. If you donwload
from QPM, it is mobile SDK, Auto QNN SDK is download from chipcode.
you can using -t to specify compilier. Normally, it is not necessary.
You can generate x86 libs , then use qnn-context-binary-generator
to generate bin format. This bin format will load into HTP/NPU direcly, it do not depend on ARM side OS.