I am using meta-tegra to build a custom image for my NVIDIA Jetson Nano. I need PyTorch, but there is no recipe for it. I built PyTorch on the device and packaged it into a wheel on the device. Now, I want to integrate that wheel into my custom Yocto layer. My image includes python3-pip and I can launch pip3 on my device. The problem is that when I try to execute my bitbake recipe, Bitbake always fails with the following error:
/home/ubuntu/Desktop/comp-jetson-yocto/build/tmp/work/jetson_nano-poky-linux/torch/1.1.0-r0/temp/run.do_install.16557: 115: /home/ubuntu/Desktop/comp-jetson-yocto/build/tmp/work/jetson_nano-poky-linux/torch/1.1.0-r0/temp/run.do_install.16557: pip3: not found
Here is my recipe:
DESCRIPTION = "NVIDIA's version of Python Torch"
HOMEPAGE = "https://nvidia.com"
LICENSE = "BSD-3-Clause"
inherit setuptools3
RDEPENDS_${PN} = "python3 \
python3-pip \
"
SRC_URI = "\
file://torch-1.1.0-cp37-cp37m-linux_aarch64.whl \
file://LICENSE \
"
SRC_URI[md5sum] = "9ec85425a64ca266abbfdeddbe92fb18"
SRC_URI[sha256sum] = "3b9b8f944962aaf550460409e9455d6d6b86083510b985306a8012d01d730b8b"
COMPATIBLE_MACHINE = "(tegra)"
COMPATIBLE_MACHINE_comp = "(-)"
LIC_FILES_CHKSUM = "file://../LICENSE;md5=acf4d595f99e159bf31797aa872aef57"
S = "${WORKDIR}/${PN}-${PV}"
do_configure() {
:
}
do_compile() {
:
}
do_install() {
pip3 install torch-1.1.0-cp37-cp37m-linux_aarch64.whl
}
PACKAGE_ARCH = "jetson_nano"
I have not been able to locate anything useful on Google on how to have my recipe install a custom wheel file with pip. How can I accomplish this? Thank you.
Bitbake needs native version of pip3
:
DEPENDS = "python3-pip-native"