Dependency Problem

[li-dahua]

Hi, I found that there is a dependency problem in the install. Your code need numpy 1, but for salesforce-lavis 1.0,2 needs numpy2.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
gensim 4.3.3 requires numpy<2.0,>=1.18.5, but you have numpy 2.0.2 which is incompatible.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
blis 1.0.1 requires numpy<3.0.0,>=2.0.0, but you have numpy 1.26.4 which is incompatible.
thinc 8.3.2 requires numpy<2.1.0,>=2.0.0; python_version >= "3.9", but you have numpy 1.26.4 which is incompatible.

[altndrr]

Hi, it looks like salesforce-lavis is the cause of all the problems. Since they don't have a clear release policy for their package, they may have updated some dependencies without changing the version.

I didn't check all the configurations exhaustively, but if you manually install the packages in the pyproject.toml file, you can solve the problem.

You can try by running the following commands:

# create the virtual environment
python -m venv .venv

# source the environment
source .venv/bin/activate

# install the dependencies in pyproject (except for salesforce-lavis) with pip
pip install "torch == 2.0.1" "torchvision == 0.15.2" "lightning == 2.0.2" "torchmetrics == 0.11.0" "hydra-core>1,<2" "hydra-colorlog>=1,<2" "hydra-optuna-sweeper>=1,<2" "wandb>=0,<1" "open_clip_torch>=2.20,<2.25" "transformers>=4,<5" "faiss-cpu>=1,<2" "flair>=0,<1" "gdown>=4,<5" "inflect>=6,<7" "kaggle>=1,<2" "lovely-tensors>=0,<1" "nltk>=3,<4" "pandas>=2,<3" "pyrootutils>=1,<2" "rich>=13,<14" "scikit-learn>=1,<2"

# install salesforce-lavis
pip install salesforce-lavis

# if needed, re-install numpy 1
pip install "numpy<2"

Let me know if that solves the problem.