Segmentation fault torch-2.0.0+rocm5.4.2 - AMD GPU #11432
Comments
|
👋 Hello @guiaugustoga987, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution. If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it. If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results. RequirementsPython>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. To get started: git clone https://github.com/ultralytics/yolov5 # clone
cd yolov5
pip install -r requirements.txt # installEnvironmentsYOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):
Status
If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit. Introducing YOLOv8 🚀We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀! Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects. Check out our YOLOv8 Docs for details and get started with: pip install ultralytics |
|
I do not think that YoloV5 supports PyTorch 2.0 (yet), but perhaps @glenn-jocher can advice on that. I have a RX 6800XT, and last time I used it to run YoloV5, I compiled PyTorch 1.13 from source because the ROCM dependencies arent properly integrated in the PIP wheel. Following the building from source instructions however made everything work properly. Now my RX 6800XT ran things at the same speed as my now ancient Geforce GTX1080 except with its larger memory it can handle a larger batch size... not sure using a RX 5600XT is worth it at all to be honest. |
|
@guiaugustoga987, you are correct that YOLOv5 does not currently support PyTorch 2.0. As for running YOLOv5 on your AMD Radeon RX 5600 XT, PyTorch with ROCm compatibility may require additional configuration on your system. However, as suggested by @th_hoff, you can try building PyTorch from source to ensure that you have the necessary dependencies. You can find instructions for building PyTorch from source at https://github.com/ROCmSoftwarePlatform/pytorch#install-pytorch-from-source. Regarding the performance on an RX 5600XT, we recommend benchmarking YOLOv5 on your hardware with various batch sizes to determine optimal performance, as it may depend on your individual use case. |
|
Thanks for the response. I tried to compile PyTorch 1.13 from source, but unexpectedly it froze my PC during installation (python setup.py develop). So, I tried to install PyTorch 1.13 from pip ( However, I used detect.py with a custom model I trained on google colab to detect the objects on a video. It worked just fine, as in colab. But it took around 1 minute after Model summary line to start the inference. Here is the code used :
Output : Is the delay normal ? After that everything seems fine as it took 5.7ms for each frame. |
|
Yes, it is "normal" unfortunally, due to the aforementioned issue with the PIP wheel. To not have the delay, you need to compile PyTorch + ROCM from source. Sorry, its a bit of a pain in the backside. The cause of the delay is this:
PyTorch PIP wheels do not bundle this database file correctly, so it is not used even if it is present on your system (in my case it is gfx1036). See my issue here: ROCm/MIOpen#1709 Note I recently had to close the issue as it needs to be adressed by PyTorch itself. But currently I do not have a Linux installation on my AMD GPU PC so I have not proceeded as of now. There is also not much Ultralytics can do here as it is completly on AMD/PyTorch. |
|
After compiling pytorch and torchvision from source I could get rid of the delay. Thank you for your suport. |
|
@guiaugustoga987 great to hear that compiling PyTorch and torchvision from source helped to get rid of the delay! It is a bit of a hassle to have to do that, but it is worth it to achieve better performance. Thank you for sharing your experience with us and please let us know if you have any further questions or concerns. We are always happy to help! |
|
👋 Hello there! We wanted to give you a friendly reminder that this issue has not had any recent activity and may be closed soon, but don't worry - you can always reopen it if needed. If you still have any questions or concerns, please feel free to let us know how we can help. For additional resources and information, please see the links below:
Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed! Thank you for your contributions to YOLO 🚀 and Vision AI ⭐ |
Search before asking
YOLOv5 Component
No response
Bug
Hello,
I'm currently trying to use yolov5 on my AMD GPU (AMD Radeon RX 5600 XT). I followed the commands :
I also installed pytorch with ROCm compatibility as instructed on :
https://pytorch.org/
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2
But whenever I try to run detect.py I get the following output :
Is there a way to use yolov5 on my AMD GPU ? Thanks in advance.
Environment
YOLOv5 🚀 v7.0-154-gf3ee596 Python-3.8.10 torch-2.0.0+rocm5.4.2 CUDA:0 (AMD Radeon RX 5600 XT, 6128MiB)
Python : 3.8.10
OS : Ubuntu 20.04
Minimal Reproducible Example
$ python3 detect.py --source ./data/images/zidane.jpg --weights yolov5s.pt --conf 0.4Additional
No response
Are you willing to submit a PR?
The text was updated successfully, but these errors were encountered: