卢旭辉
Perhaps you have specified A type runtime in yml file to convert model but specified B type runtime in yml file to run model
@qdushuaicao 可以在ndk-depends找不到的话可以在 /opt/android-ndk-r16c/中拷贝一个。 DSP的问题比较麻烦，不是需要root，是需要签名。
@madhu-korada Yes MACE transfers all of the weights of NN completely to GPU memory while invoking the MACE engine. You can benchmark the op's latency as follows: `python tools/converter.py run...
@umairjavaid MACE does not support nvidia gpu, for it need OpenCL
@leiwen83 Could you use the master branch? we now work on the master, thank you first
@leiwen83 Sorry for the late reply, please use `bazel` command, for example: `python tools/bazel_adb_run.py --target="//test/ccbenchmark:mace_cc_benchmark" --run_target=True --target_abis=armeabi-v7a,arm64-v8a`
@sumant85 I'm sorry to reply so late. It is a good suggestion, recently we are considering at refactoring the framework, we'll support this feature based on the new framework.
It is supported but we remove the hta's code because we are not authorized by Qualcomm.
@luqiang6q 方便上传一下你的训练代码么？
@luqiang6q I'm very sorry for the late reply. Those days I was very busy and forgot this problem. Now we have resolved this problem, please use the version v1.0.4.