Yolo v5 loading using torch 2.6

[abcjoh]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Code
model = torch.hub.load("ultralytics/yolov5:v7.0", 'custom', model_path, force_reload=True)

Error
/torch/hub/ultralytics_yolov5_v7.0/hubconf.py", line 60, in _create
model = attempt_load(path, device=device, fuse=False) # arbitrary model
File "/home/.cache/torch/hub/ultralytics_yolov5_v7.0/models/experimental.py", line 79, in attempt_load
ckpt = torch.load(attempt_download(w), map_location='cpu') # load
File "/home/venv/lib/python3.10/site-packages/torch/serialization.py", line 1470, in load
raise pickle.UnpicklingError(_get_wo_message(str(e))) from None
_pickle.UnpicklingError: Weights only load failed. This file can still be loaded, to do so you have two options, �[1mdo those steps only if you trust the source of the checkpoint�[0m.
(1) In PyTorch 2.6, we changed the default value of the weights_only argument in torch.load from False to True. Re-running torch.load with weights_only set to False will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source.
(2) Alternatively, to load with weights_only=True please check the recommended steps in the following error message.
WeightsUnpickler error: Unsupported global: GLOBAL models.yolo.DetectionModel was not an allowed global by default. Please use torch.serialization.add_safe_globals([DetectionModel]) or the torch.serialization.safe_globals([DetectionModel]) context manager to allowlist this global if you trust this class/function.

Check the documentation of torch.load to learn more about types accepted by default with weights_only https://pytorch.org/docs/stable/generated/torch.load.html.

Question
Can you please help me to resolve this I have tried to also add the weights_only = False but it is not working. Torch version is 2.6.
Is yolov5 not compatible with this torch 2.6 version or am I missing something.

Additional

No response

[UltralyticsAssistant]

👋 Hello @abcjoh, thank you for your interest in YOLOv5 🚀 and for reporting this issue! Let's get this resolved. Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for tasks like Custom Model Loading and more.

From your description, it appears the issue may be related to 🐛 loading YOLOv5 weights with PyTorch 2.6. To help debug this, could you kindly provide a minimum reproducible example (MRE)? For example:

• The exact steps or script you’re using to reproduce the issue
• Any model weights or trained checkpoints (if possible, without sensitive/private data)
• The logs of errors you're encountering and the command used to run your script

This will help us investigate further.

Requirements

YOLOv5 requires Python>=3.8.0 with all requirements.txt installed, including PyTorch>=1.8. Please verify your environment setup:

git clone https://github.com/ultralytics/yolov5  # clone repo
cd yolov5
pip install -r requirements.txt  # install packages

For compatibility updates with new PyTorch versions (like 2.6), verifying all dependencies within an environment from scratch may help.

Quickstart Environments 🌟

YOLOv5 can also be run in any of these up-to-date verified environments for hassle-free setups:

• Notebooks with free GPU:
• Google Cloud: Deep Learning VM Quickstart Guide
• Amazon AWS: Deep Learning AMI Guide
• Docker Image: Docker Setup Guide

Documentation

YOLOv5 relies on the torch.load method for model weights loading. In your case, the changes in PyTorch 2.6 (such as the weights_only default behavior) seem to trigger the error. Ensuring up-to-date compatibility or downgrading PyTorch to 2.5 might temporarily resolve your issue 🤔.

👉 Pro Tip: Try adding torch.serialization.add_safe_globals([DetectionModel]) in your script before loading the model, as suggested by the PyTorch error message. This explicitly allows the model class for serialization.

An Ultralytics engineer will take a closer look at this soon. Thank you for your patience! 😊

[pderrenger]

To resolve the PyTorch 2.6 compatibility issue with YOLOv5, please ensure you're using the latest YOLOv5 code from the master branch (not the v7.0 release tag). Update with:

pip install -U git+https://github.com/ultralytics/yolov5.git

Then try loading your model again. This update includes compatibility improvements for newer PyTorch versions. If issues persist, confirm your model file is uncorrupted and from a trusted source.

For reference, YOLOv5 actively maintains PyTorch compatibility as shown in our release notes: v7.0 updates.