Open rezaghoddoosian opened 5 months ago
It seems that this project has many bugs. For example, the following dependencies are missed: dmvr, clip and official. You can install them using the following code:
pip install git+https://github.com/deepmind/dmvr.git
conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0
pip install ftfy regex tqdm
pip install git+https://github.com/openai/CLIP.git
pip install -U -q "tf-models-official"
Still, some other parts of the code need to be improved.
Hi Masoud, Yeah I noticed. I installed some of these. My biggest issue is the system requirements. I have CUDA 11.3, so if I need to install the corresponding jaxlib for my CUDA, I wonder if this code works for CUDA 11.3? Were you able to run this code eventually?
Hi Reza, I run the code on Colab. Although after installing dependencies the problem of requirements has been solved, it still has many problems that need to be fixed. (The code is not clean enough and needs so much improvement!) I finally gave up. I hope developers clean up the code and provide some comments.
Hi Masoud, 1-Can you elaborate what you mean by "it has many problems that need to be fixed"? what specific problems? 2- What version of jaxlib and Cuda did you use?
Hi Reza,
I suggest you first run the code on Colab. When you use Colab you have no problem with versions of jaxlib and you could be sure that the code is clean. When you are sure about the code itself, then you can run it locally. I did not run the code on a local machine, so I cannot tell you about versions difference.
My question is regarding the implementation of this paper "Verbs in Action: Improving verb understanding in video-language models" (verbs_in_action). Im having trouble figuring out the dependencies to run this code. Can this scenic code (verbs_in_action) run with jax for CUDA 11.3? Also, wondering where I can access the requirements for this implementation?