Hi, all:
I noticed https://docs.ultralytics.com/guides/nvidia-jetson/#install-ultralytics-package, torch and torchvision now only support JP6.0. However, JP6.1 is already released, please see https://forums.developer.nvidia.com/t/jetpack-6-1-release-announcement/308185 . Can anybody help to build the most up-to-date torch and torchvision for JP6.1?
Cheers
I would recommend checking the Jetson Dockerfile in the Ultralytics repo for straightforward setup. AFAIK, this should work for any Jetpack version, but to be honest, I’ve never set up a Jetson device personally, you’ll likely need to modify this line in the Dockerfile:
- FROM nvcr.io/nvidia/l4t-jetpack:r36.3.0
+ FROM nvcr.io/nvidia/l4t-jetpack:r36.4.0
Since r36.4.0 has Jetpack 6.1 according to this NVIDIA page
Otherwise, see the NVIDIA Jetson Download Center for the latest PyTorch version 2.5.0 but doesn’t look like they have a wheel for torchvision (yet).
➜ docker git:(main) sudo docker build -f Dockerfile-jetson-jetpack6 -t pytorch-jetson-jetpack6 .
DEPRECATED: The legacy builder is deprecated and will be removed in a future release.
Install the buildx component to build images with BuildKit:
https://docs.docker.com/go/buildx/
Sending build context to Docker daemon 32.77kB
Step 1/13 : FROM nvcr.io/nvidia/l4t-jetpack:r36.3.0
---> b859f440ba7f
Step 2/13 : ENV PYTHONUNBUFFERED=1 PYTHONDONTWRITEBYTECODE=1 PIP_NO_CACHE_DIR=1 PIP_BREAK_SYSTEM_PACKAGES=1
---> Using cache
---> 64666bd99085
Step 3/13 : ADD https://github.com/ultralytics/assets/releases/download/v0.0.0/Arial.ttf https://github.com/ultralytics/assets/releases/download/v0.0.0/Arial.Unicode.ttf /root/.config/Ultralytics/
Downloading [==================================================>] 773.2kB/773.2kB
Downloading [==================================================>] 23.28MB/23.28MB
---> Using cache
---> 35b818fd7135
Step 4/13 : RUN apt-get update && apt-get install -y --no-install-recommends git python3-pip libopenmpi-dev libopenblas-base libomp-dev && rm -rf /var/lib/apt/lists/*
---> Using cache
---> 22ac1844de4a
Step 5/13 : WORKDIR /ultralytics
---> Using cache
---> 97f964c29a89
Step 6/13 : COPY . .
---> Using cache
---> 8592eebb7d1b
Step 7/13 : RUN sed -i '/^\[http "https:\/\/github\.com\/"\]/,+1d' .git/config
---> Running in 5eac5d2a3dc9
sed: can't read .git/config: No such file or directory
The command '/bin/sh -c sed -i '/^\[http "https:\/\/github\.com\/"\]/,+1d' .git/config' returned a non-zero code: 2
2.5.0. However, when I ran it, I got:➜ python
Python 3.10.12 (main, Sep 11 2024, 15:47:36) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/local/lib/python3.10/dist-packages/torch/__init__.py", line 361, in <module>
from torch._C import * # noqa: F403
ImportError: libcusparseLt.so.0: cannot open shared object file: No such file or directory
>>>
Thank you @BurhanQ .
I need to overwrite a lot of things, including both the PyTorch located in NVIDIA Jetson Download Center and TorchVision … a lot more actually. Now I have
And, tried YOLO11 as well… Thank you
@jiapei100 glad to hear you got it working! It would be great if you could share the changes you made here or even better, open a PR to include those in the GitHub repo for others.
Glad to hear everything is working now @jiapei100. As @BurhanQ mentioned, if you want to update to JetPack6.1 with Docker support, please do the changes. I am currently working on it and will update here once I made it officially available.
FYI: Since JetPack6.1 comes with TensorRT 10.3, you have access to latest TensorRT as well.
Also, for PyTorch and Torchvision, you can look into this website. You will find the latest .whl files there. This is managed by Dustin Franklin from NVIDIA who is famous for dusty-nv/jetson-containers.
So I may update the new JetPack6.1 with these new PyTorch and Torchvision versions as well.
Thanks for your patience.
I came across the same problem with Jetson AGX / Jetpack 6.1
I used torch from Dustin’s Site.
wget http://jetson.webredirect.org/jp6/cu126/+f/5cf/9ed17e35cb752/torch-2.5.0-cp310-cp310-linux_aarch64.whl#sha256=5cf9ed17e35cb7523812aeda9e7d6353c437048c5a6df1dc6617650333049092
pip install torch-2.5.0-cp310-cp310-linux_aarch64.whl
then installed them respective torchvision from Dustin’s site.
wget http://jetson.webredirect.org/jp6/cu126/+f/5f9/67f920de3953f/torchvision-0.20.0-cp310-cp310-linux_aarch64.whl#sha256=5f967f920de3953f2a39d95154b1feffd5ccc06b4589e51540dc070021a9adb9
pip install torchvision-0.20.0-cp310-cp310-linux_aarch64.wh
then installed onnxruntime 1.20 from justin s site.
wget http://jetson.webredirect.org/jp6/cu126/+f/0c4/18beb3326027d/onnxruntime_gpu-1.20.0-cp310-cp310-linux_aarch64.whl#sha256=0c418beb3326027d83acc283372ae42ebe9df12f71c3a8c2e9743a4e323443a4
pip install onnxruntime_gpu-1.20.0-cp310-cp310-linux_aarch64.whl
Downgraded the numpy as usual.
and lastly installed libcusparselt0 from nvidia site.
Hello @jiapei100 We have updated with JetPack6.1 which includes TensorRT 10.3 support with Docker:
This will do the job!
t=ultralytics/ultralytics:latest-jetson-jetpack6
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t
Thanks for waiting!
CC: @BurhanQ
Downgraded to which version of numpy i am facing error with c-api in torch
Hi @Venkat1405,
It should be numpy 1.23.5.
We are already handling that here when you do pip install ultralytics[export] on the Jetson: