I'm experiencing several issues while trying to install LAVIS.
At first, I believed the possible problems were related to GPU support (RTX 5060), but I realized that the difficulties arise much earlier — already during the basic installation process.
I followed exactly the steps described in the official GitHub repository:
conda create -n lavis python=3.8
conda activate lavis
pip install salesforce-lavis
git clone https://github.com/salesforce/LAVIS.git
cd LAVIS
pip install -e .
However, during the installation, several dependency conflicts and library incompatibilities prevented LAVIS from working properly:
moviepy >= 1.25.0
scipy (current) >= 1.25.0
salesforce-lavis (via opencv-python-headless==4.5.5.64) <= 1.24.x
spaces (used by interfaces like Gradio) >= 2.0.0
In other words, if I install using Python 3.8, I can't use moviepy version 2.0.0.
If I try with Python 3.9, I need opencv in a lower version, which also causes conflicts.
As far as I can remember, the main issues involve these libraries or something related.
The point is that when I use Python 3.8, it asks for libraries compatible with Python 3.9 — and when I use Python 3.9, it asks for libraries compatible with Python 3.8.
At this point, I don’t have the technical knowledge to resolve these issues on my own.
I believe I would need to know the exact version of each library to install, but I don’t know how to find out what those versions should be.
Once I overcome these dependency conflicts, there is still a separate issue that remains: GPU compatibility.
Even if I manage to solve the installation issues above, I will likely face problems related to my graphics card. Maybe the command below would work — or it could introduce new library conflicts:
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129
My system configuration:
Operating System: Windows 11
GPU: NVIDIA GeForce RTX 5060
I'm experiencing several issues while trying to install LAVIS.
At first, I believed the possible problems were related to GPU support (RTX 5060), but I realized that the difficulties arise much earlier — already during the basic installation process.
I followed exactly the steps described in the official GitHub repository:
conda create -n lavis python=3.8conda activate lavispip install salesforce-lavisgit clone https://github.com/salesforce/LAVIS.gitcd LAVISpip install -e .However, during the installation, several dependency conflicts and library incompatibilities prevented LAVIS from working properly:
moviepy >= 1.25.0
scipy (current) >= 1.25.0
salesforce-lavis (via opencv-python-headless==4.5.5.64) <= 1.24.x
spaces (used by interfaces like Gradio) >= 2.0.0
In other words, if I install using Python 3.8, I can't use moviepy version 2.0.0.
If I try with Python 3.9, I need opencv in a lower version, which also causes conflicts.
As far as I can remember, the main issues involve these libraries or something related.
The point is that when I use Python 3.8, it asks for libraries compatible with Python 3.9 — and when I use Python 3.9, it asks for libraries compatible with Python 3.8.
At this point, I don’t have the technical knowledge to resolve these issues on my own.
I believe I would need to know the exact version of each library to install, but I don’t know how to find out what those versions should be.
Once I overcome these dependency conflicts, there is still a separate issue that remains: GPU compatibility.
Even if I manage to solve the installation issues above, I will likely face problems related to my graphics card. Maybe the command below would work — or it could introduce new library conflicts:
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129My system configuration:
Operating System: Windows 11
GPU: NVIDIA GeForce RTX 5060