我们正在使用骁龙神经处理引擎和 QNN, 正在尝试验证我们的 ML 模型是否在此处理器上运行。我们使用 QCS8550 LE 并收到以下错误:

[weixin_32291731]

我们正在使用骁龙神经处理引擎和 QNN, 正在尝试验证我们的 ML 模型是否在此处理器上运行。我们使用 QCS8550 LE  并收到以下错误:
root@kalama:~# snpe-throughput-net-run --verbose --container detection.dlc --duration 60 --perf_profile balanced --use_gpu
Before SNPE setup, allocated native heap is 1708448
Before SNPE setup, allocated mmap region(s) is 532480
The selected runtime gpu_float32_16_hybrid is not available on this platform. Continue anyway to observe the failure at network creation time.
[Instance 0 - gpu_float32_16_hybrid] Unable to successfully create SNPE instance. Will continue to creation of other SNPE instance(s).container=detection.dlc; perf_profile=balanced; error_code=1002; error_message=Layer parameter value is invalid. No backend could validate Op=input.nhwc Type=Transpose error code=3110; error_component=Model Validation; line_no=131; thread_id=548616069152
Error: Requested 1 instances of SNPE, Could not create any.. Exiting.

root@kalama:~# snpe-throughput-net-run --verbose --container detection.dlc --duration 60 --perf_profile balanced --use_cpu
Before SNPE setup, allocated native heap is 1708448
Before SNPE setup, allocated mmap region(s) is 532480
The selected runtime cpu_float32 is not available on this platform. Continue anyway to observe the failure at network creation time.
[Instance 0 - cpu_float32] Unable to successfully create SNPE instance. Will continue to creation of other SNPE instance(s).container=detection.dlc; perf_profile=balanced; error_code=1002; error_message=Layer parameter value is invalid. No backend could validate Op=input.nhwc Type=Transpose error code=3110; error_component=Model Validation; line_no=131; thread_id=547621818400
Error: Requested 1 instances of SNPE, Could not create any.. Exiting.

root@kalama:~# snpe-throughput-net-run --verbose --container detection.dlc --duration 60 --perf_profile balanced --use_dsp
Before SNPE setup, allocated native heap is 1708448
Before SNPE setup, allocated mmap region(s) is 532480
The selected runtime dsp_fixed8_tf is not available on this platform. Continue anyway to observe the failure at network creation time.
[Instance 0 - dsp_fixed8_tf] Unable to successfully create SNPE instance. Will continue to creation of other SNPE instance(s).container=detection.dlc; perf_profile=balanced; error_code=1002; error_message=Layer parameter value is invalid. No backend could validate Op=input.nhwc Type=Transpose error code=3110; error_component=Model Validation; line_no=131; thread_id=548176212000
Error: Requested 1 instances of SNPE, Could not create any.. Exiting.
root@kalama:~#@

[weixin_38498942]

从错误消息来看，您似乎有一个将输入转置为 NHWC 格式的转置层，并且任何运行时都不支持此层。这很奇怪，因为转置操作应该受 SNPE支持。此外，奇怪的是您的设备上没有任何运行时可用。在 CQS8550 LE 1.0 基线上应该同时具有 CPU 和 DSP 运行时，由于已知问题，只有GPU 运行时现在无法正常工作。您是否在设备上正确设置了 SNPE 环境？

例如将对应的snpe的库, dsp的库放到 /usr/lib 和/usr/lib/rfsa/adsp/ 下, 然后再尝试 用 snpe-platform-validator 工具验证您的环境，这样应该能够正常的运行DSP.