I am using the Jetson Nano Super Developer Kit/Jetson JetPack 6.2.2. with python 3.10 I am using openwake, whisper and pipertts. I created a onnxruntime build to attempt to use piper with the gpu (that build took 6 hours to complete)- works but I want it to be faster - too make more of a social-animatronic - I am not using a local LLM as it was too slow. FYI tried the same thing on the new Raspberry PI with 16gb ram and hated it. - Build that was auccessful was - cd /home/robit/Documents/onnxruntime
ls -l build.sh # sanity check it exists
./build.sh
–build_dir /home/robit/Documents/ort_build/Linux
–config Release
–update --build
–parallel 1
–build_wheel
–use_cuda
–cuda_home /usr/local/cuda
–cudnn_home /usr/lib/aarch64-linux-gnu
–skip_tests
–cmake_extra_defines
CMAKE_CUDA_ARCHITECTURES=87
onnxruntime_USE_CUDA_RNN=OFF
onnxruntime_NVCC_THREADS=1
onnxruntime_ENABLE_CPU_FP16_OPS=OFF
onnxruntime_USE_FLASH_ATTENTION=OFF
onnxruntime_USE_MEMORY_EFFICIENT_ATTENTION=OFF
CMAKE_POLICY_VERSION_MINIMUM=3.5
CMAKE_CXX_STANDARD=20
CMAKE_CXX_STANDARD_REQUIRED=ON
ABSL_PROPAGATE_CXX_STD=ON
onnxruntime_USE_PREINSTALLED_EIGEN=ON
eigen_SOURCE_PATH=/home/robit/Documents/deps/eigen/eigen-e7248