NVIDIA / modulus-launch

Repo of optimized training recipes for accelerating PyTorch workflows of AI driven surrogates for physical systems
Apache License 2.0
54 stars 27 forks source link

🐛[BUG]: ModuleNotFoundError: No module named 'modulus.launch' #96

Open SamiraMalek opened 1 year ago

SamiraMalek commented 1 year ago

Version

0.2.0

On which installation method(s) does this occur?

Pip

Describe the issue

Hello,

I've installed "nvidia-modulus" and successfully cloned the repository named "modulus-launch-0.2.0". However, I encountered an issue when attempting to run the "train.py" script located at the path "modulus-launch-0.2.0/examples/cfd/ahmed_body_mgn/train.py". The specific error message I received is: "ModuleNotFoundError: No module named 'modulus.launch'".

I would greatly appreciate it if you could guide me on how to resolve this issue.

Minimum reproducible example

No response

Relevant log output

No response

Environment details

No response

mnabian commented 1 year ago

Hi @SamiraMalek , you also need to install modulus-launch, either through PyPi (https://github.com/NVIDIA/modulus-launch#pypi), or by running pip install . in your cloned modulus-launch folder. Please try this and let us know if you are still having issues. Thanks for choosing Modulus!

SamiraMalek commented 1 year ago

Hi @mnabian ,

I greatly appreciate your swift response. I've successfully set up modulus-launch, yet I encountered the subsequent error message: PydanticUserError - In case you utilize @root_validator with pre=False (which is the default), it is IMPERATIVE to indicate skip_on_failure=True. It's worth noting that @root_validator is outdated and should be substituted with @model_validator.

mnabian commented 1 year ago

@SamiraMalek could you please share the full logs?

SamiraMalek commented 1 year ago

Hi,

Certainly, You can access all the logs by following this URL link: https://colab.research.google.com/drive/1ay1GHT5uBr6LZGbu9keejsCTTJ0RLPSJ?usp=sharing Within this notebook, you'll find a total of 8 cells, structured as outlined below:

  1. Cloning modulus-launch repository
  2. Changing directory to where modulus-launch is cloned
  3. Installing nvidia-modulus.launch
  4. Installing required packages for modulus-launch/examples/cfd/ahmed_body_mgn/train.py 4-1. pip install hydra-core 4-2. pip install dgl 4-3. pip install vtk 4-4. pip install pyvista
  5. Running modulus-launch/examples/cfd/ahmed_body_mgn/train.py

I duplicated these steps using "nvidia-modulus" version 0.2.0, sourced from: https://github.com/NVIDIA/modulus-launch/archive/refs/tags/v0.2.0.zip. However, I opted to download the aforementioned zip file instead of performing step 1 ("Cloning the modulus-launch repository"). I then proceeded to upload the extracted files to my Google Drive and subsequently linked my Google Drive to Google Colab. For a comprehensive record of this process, please refer to the detailed log available at this link: https://colab.research.google.com/drive/1r0O6c7A02aqbhgntlWrJBdjm2eLUc7XJ?usp=sharing

If you need more information, please let me know.

mnabian commented 1 year ago

Thanks for providing the notebook. The issue is with the latest version of pydantic. I changed it to use pydantic==1.10.2 and it works now. Please check and verify. Note that for this example you need to have the dataset, which is not released yet. Also note that you were installing from pip wheel and not from source. You mentioned you tried using version 0.2.0 from source, but you were actually installing from the latest wheel.

SamiraMalek commented 1 year ago

@mnabian Thank you so much for your help. I meant files on the link for version 0.2.0, I did not mean installing from source. Can I use the pre-trained model?

mnabian commented 1 year ago

The pre-trained model will be released in a few days, along with a few samples from the dataset. I'll update you once these are available.

mnabian commented 1 year ago

Hi @SamiraMalek , the pre-trained model and a sample dataset are published here: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/modulus/models/modulus_ahmed_body_meshgraphnet, https://catalog.ngc.nvidia.com/orgs/nvidia/teams/modulus/resources/modulus_datasets-ahmed_body_test.

Please let me know if you have any other issues with installing modulus-launch, or else let us close this issue.

SamiraMalek commented 1 year ago

Hello @mnabian,

I wanted to express my gratitude for your assistance. I'm wondering if it would be feasible for you to upload the nvidia-modulus and nvidia-modulus.launch packages to the Anaconda repository. This step would enable the installation of nvidia-modulus.launch via conda. Currently, when I attempt to execute the "conda install nvidia-modulus.launch" command, I encounter the following error message: PackagesNotFoundError.

Warm regards, Samira

mnabian commented 1 year ago

Unfortunately, there are no plans for publishing Modulus on conda but we can revisit this in the future.